Rapid recognition and lithotripsy parameter adaptive matching method in urinary calculus component operation

Abstract

The invention discloses a rapid recognition and lithotripsy parameter adaptive matching method in a urinary calculus component operation, and relates to the technical field of medical detection and urinary surgical operations. A stone component-lithotripsy parameter database is constructed before an operation, and spectrum and ultrasonic characteristics and corresponding optimal lithotripsy parameters are collected; stone multi-modal data are synchronously collected through the integrated probe in the operation; preprocessing the data and extracting feature vectors; inputting the feature vectors into an improved deep residual network to quickly identify components; calling an initial parameter based on an identification result, and correcting the parameter through a component-hardness model; monitoring the fragmentation state in real time during the operation and optimizing parameters; after the operation, recording data and updating the database, and adopting incremental learning to improve the subsequent accuracy. According to the method, the stone component recognition accuracy and real-time performance are improved, self-adaptive matching of stone crushing parameters is achieved, complications are reduced, the operation efficiency is improved, and continuous optimization can be achieved through data accumulation to adapt to various clinical requirements.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical detection and urological surgery, and particularly relates to a method for quickly identifying and adaptively matching lithotripsy parameters in urological stone composition surgery. BACKGROUND

[0002] Urological stones are common diseases in urology, and extracorporeal shock wave lithotripsy (ESWL) has become the preferred treatment method in clinic due to its small trauma and fast recovery. The key to successful surgery is to accurately identify the composition of the stone and match the appropriate lithotripsy parameters. However, in the current clinical process, there is a clear bottleneck in the identification of stone composition and the matching of lithotripsy parameters. Although CT and infrared spectroscopy can be used to preliminarily determine the composition of the stone before surgery, CT can only roughly classify according to density and cannot distinguish between stones with similar compositions, such as calcium oxalate and calcium phosphate. Moreover, the preoperative detection samples are mostly previous stone fragments or puncture samples, which cannot fully reflect the composition distribution of the actual stone during surgery, especially for mixed composition stones. The error rate of preoperative judgment is relatively high. Although infrared spectroscopy can accurately identify the composition, it requires an ex vivo sample and takes a long time to detect, which cannot be applied to real-time decision-making during surgery, resulting in the need to rely on the experience of the surgeon to determine the composition of the stone during surgery, which is highly subjective and limited in accuracy.

[0003] The shortcomings of intraoperative stone composition identification technology further exacerbate the treatment dilemma. Existing intraoperative rapid identification methods mostly use a single detection modality, such as high-frequency ultrasound or near-infrared spectroscopy. Ultrasound detection relies on the acoustic impedance difference of the stone, but the echo signal is easily distorted due to the influence of blood and tissue obstruction and the reflection of the stone surface during surgery, making it difficult to extract clear features. Near-infrared spectroscopy can distinguish between components through characteristic absorption peaks, but the exposed area of the stone is limited during surgery, and the coupling stability between the probe and the stone is poor, which easily leads to incomplete spectral data due to slight displacement. The recognition accuracy is generally low. More importantly, the detection time of these single modality methods usually exceeds 2 seconds, which does not meet the needs of intraoperative rapid decision-making, which may prolong the operation time and increase the risk of anesthesia for patients.

[0004] The blindness of stone parameter matching directly affects the treatment effect and safety. At present, the fixed stone parameters are set based on the experience of doctors in clinic, such as uniformly using 3J energy and 10Hz frequency for impact, without considering the physical property difference caused by the composition difference of stones: the hardness of calcium oxalate stone is high, and the brittleness is low, so that the fixed parameters may cause incomplete fragmentation due to insufficient energy, and repeated impact is needed; the hardness of uric acid stone is low, and the brittleness is high, so that the over-high energy may easily cause excessive fragmentation of the stone, the fragments are small and distributed scattered, the difficulty of stone removal is increased, and even the surrounding mucosa tissue may be damaged by the excessive impact of the shock wave, and complications such as hematuria and edema are caused. At the same time, the real-time monitoring of the fragmentation state of the stone is lacked during the operation, the parameters cannot be dynamically adjusted according to the particle size and distribution density of the fragments, and the effect can only be judged by relying on the postoperative imaging examination, so that the treatment efficiency is low and there is a safety hazard. In summary, an integrated method for real-time and accurate identification of the composition of urinary stones and automatic matching of the optimal stone parameters during the operation is urgently needed in clinic, so as to break through the limitations of the existing technology and improve the safety and effectiveness of the urinary stone fragmentation operation. SUMMARY

[0005] The urinary stone composition intraoperative rapid identification and stone parameter self-adaptive matching method provided by the present application solves the problems mentioned in the prior art.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: a urinary stone composition intraoperative rapid identification and stone parameter self-adaptive matching method, comprising:

[0007] Preoperative construction of stone composition-stone parameter database: collect the spectral feature library and ultrasonic echo feature library of common stone compositions, the spectral feature library covers the characteristic absorption peaks of calcium oxalate, calcium phosphate, uric acid, cystine and mixed stones in the 350-2500nm wave band, the ultrasonic echo feature library includes the acoustic impedance, attenuation coefficient and backscattering coefficient of stones with different compositions under the frequency of 5-15MHz, and the optimal stone parameter combination of the corresponding composition is also stored, including the shock wave energy, impact frequency and focusing depth;

[0008] Real-time acquisition of intraoperative data: the stone data is synchronously acquired by the mini near-infrared spectrum probe and high-frequency ultrasonic probe integrated in the stone fragmentation device, the ultrasonic probe adopts pulse-echo mode, and the acquisition area covers the exposed surface of the stone and the diameter is greater than or equal to 2mm;

[0009] Data preprocessing and feature extraction: the spectral data is subjected to baseline correction, smoothing denoising and normalization processing, the SG filter is used to remove high-frequency noise, the window width is set to 11 points, the first derivative transformation is used to enhance the absorption peak characteristics, and the absorbance values at 5-8 characteristic wavelengths are extracted; the envelope detection and Fourier transform are performed on the ultrasonic data, the peak intensity, main peak frequency and bandwidth characteristics of the echo signal are extracted to form a feature vector;

[0010] Stone composition recognition: input the extracted feature vector into the deep residual network model, the model contains 4 residual blocks, each residual block is composed of 2 convolutional layers and jump connection, output the probability distribution of each stone composition through the Softmax function, when the highest probability is greater than or equal to 90%, directly determine the composition, when 80%-90%, combined with clinical history to assist in judgment, the determination time is less than or equal to 1s, the composition recognition accuracy is greater than or equal to 95%;

[0011] Lithotripsy parameter adaptive matching: based on the identification result, call the initial parameter combination in the database, modify the parameters through the established composition-hardness correlation model, when the hardness of calcium oxalate stones is higher, increase the initial energy by 10%-20%, when the hardness of uric acid stones is lower, reduce the energy by 15%-25%, and at the same time, dynamically adjust the focusing depth according to the diameter of the stone, increase the focusing depth by 0.5mm for every 1mm increase in diameter;

[0012] Real-time feedback and parameter optimization: during the operation, the stone fragmentation state is monitored in real time through the ultrasonic probe, the lithotripsy area image is collected every 0.5s, the average particle size and distribution density of the fragments are analyzed, when the fragment particle size is less than or equal to 2mm and the proportion is greater than or equal to 80%, reduce the impact frequency to 5-10Hz, when there is no obvious change in the fragments after continuous 3 times of impact, increase the energy by 5%-10%, until the stone fragmentation meets the standard;

[0013] Postoperative parameter recording and database updating: automatically record the stone composition, actual use of lithotripsy parameters and fragmentation effect evaluation of this operation, update the database using incremental learning algorithm, adjust the model once every 50 clinical data accumulated.

[0014] Further, it further comprises:

[0015] The confidence reinforcement step in stone composition recognition: the matching degree of the characteristic wavelength interval is calculated by integration, the formula is Where C is the matching degree of the characteristic interval, λ is the wavelength, λ1 and λ2 are the starting and ending wavelengths of the characteristic interval respectively, I(λ) is the spectral intensity of the stone to be identified, and W(λ) is the weight function of the standard spectrum, when the fusion confidence is greater than or equal to 85%, the composition is confirmed, otherwise, secondary collection is started, and the recognition reliability of mixed composition stones and small stones is improved through verification.

[0016] The fine control step of dynamic adjustment of lithotripsy energy: based on the stone fragmentation rate, the energy output is modified in real time, the energy change rate formula is Where is the energy change rate with time, k is the proportionality coefficient, the value is 0.02-0.05 and decreases with the increase of tissue distance, when the distance between the stone and the surrounding tissue is less than or equal to 2mm, k is 0.02, The volume change rate of the stone fragmentation per unit time is calculated through the ultrasound image sequence, H is the stone hardness value estimated by the composition-hardness model, and the energy output is dynamically adjusted to change with the fragmentation process in stages. The initial stage has a faster energy increasing rate, and when the fragment ratio reaches 50%, the rate is reduced by 30%, reducing the impact damage to the surrounding tissues and shortening the fragmentation time.

[0017] Further, when constructing the stone composition-lithotripsy parameter database before operation, a stratified sampling combined with a multi-center data fusion method is adopted. The samples include different regional patient groups, and the CT value of the stone, the urine pH value of the patient and the metabolic indicators are recorded synchronously. The stone samples of patients with metabolic abnormalities are marked as a special subset, and a parameter matching sub-model is trained for the subset. The database uses a distributed storage architecture to support real-time uploading of case data by medical institutions.

[0018] Further, when collecting data in real time during the operation, the spectral probe and the ultrasonic probe are coaxially integrated, and the probe front end is equipped with a telescopic water bag coupling device. The water bag is filled with 37℃ sterile normal saline, and the pressure in the bag is controlled at 5-10mmHg by a pressure sensor. The spectral acquisition uses a ring-shaped LED array light source with a power of 50-100mW and automatically adjusts according to the reflectivity of the stone surface.

[0019] Further, in the data preprocessing and feature extraction, the standard normal variable transformation is used for the normalization processing of the spectral data, and the formula is wherein is the normalized spectral intensity, I(λ) is the original spectral intensity, μ is the mean of the spectral intensity, and σ is the standard deviation. After processing, the effectiveness is verified by calculating the cosine similarity with the standard sample. If the similarity is ≥0.95, it is considered to be qualified, otherwise the abnormal data repair process is started.

[0020] Further, in the stone composition identification, the deep residual network model introduces a hybrid attention mechanism. Channel attention layers and spatial attention layers are added in sequence after each residual block. The channel attention layer learns the weight of the spectral band and the ultrasonic feature through the squeeze-excitation operation. The spatial attention layer focuses on the feature difference between the edge and the core area of the stone. The model uses a lightweight design. The model training uses a hybrid loss function composed of cross-entropy loss and center loss with a weight of 8:2. The center loss constrains the clustering of features of the same composition. The batch size is set to 32, the iteration number is 500, the initial learning rate is 0.001, and the learning rate is attenuated by 10% every 100 iterations. The early stopping strategy is used to prevent overfitting. The training is terminated when the validation set accuracy does not improve for 50 consecutive iterations.

[0021] Further, in the stone parameter adaptive matching, the component-hardness correlation model is established by multivariate nonlinear regression, and the relationship between the hardness value H and the characteristic wavelength absorbance A1, A2, ultrasonic acoustic impedance Z and attenuation coefficient alpha is Wherein a, b, c, d are regression coefficients, e is a constant term, which is fitted by 500 clinical data, and a=12, b=18, c=5, d=0.8, e=3 for calcium oxalate stones, a=10, b=15, c=4, d=0.6, e=2 for calcium phosphate stones, and a=8, b=12, c=3, d=0.5, e=1 for uric acid stones; When the parameters are matched, the stone location correction factor is introduced at the same time, the impact frequency of the stone located in the renal pelvis is reduced by 10%, and the focusing depth accuracy of the stone located in the upper ureter is improved to ±0.2mm.

[0022] Further, in the real-time feedback and parameter optimization, the calculation of the average particle size of the fragments adopts a multi-scale feature fusion method based on the ultrasonic image, the gray level co-occurrence matrix features and LBP features of 10*10, 20*20 and 30*30 pixel windows are extracted, and the fragment size is obtained by inputting the particle size estimation model; When the stone displacement is monitored to be more than 1mm, the dynamic tracking algorithm is started, the displacement vector is calculated through 5 continuous images, the motion trajectory is predicted, and the focusing depth is adjusted 50ms in advance, the trajectory prediction adopts a second-order polynomial fitting; The tissue protection threshold is set, when the distance between the shock wave focus and the mucosa is less than or equal to 1.5mm, the energy is automatically reduced by 30% and the impact interval is increased to 0.5s, so as to avoid tissue damage, and the damage risk is monitored through the elastic characteristics of the ultrasonic echo, when the elastic coefficient change rate is greater than 15%, the impact is paused and a warning is issued.

[0023] Further, in the postoperative parameter recording and database updating, the fragmentation effect evaluation adopts a fragmentation efficiency index E, which is defined as the ratio of the total mass of the fragments with a diameter of less than or equal to 2mm after fragmentation to the mass of the original stone, and the comprehensive score S=0.5·E+0.3·(1 / T)+0.2·(1 / B) is calculated by combining the operation time T and the patient's bleeding amount B, and the higher the score is, the better the effect is; Incremental learning adopts a federal learning framework, each medical institution locally trains a model, only uploads the model parameter update amount, protects the patient's privacy, and adjusts the model once every 50 cases of data, and when the model is adjusted, the knowledge distillation technology is used to migrate the knowledge of the new data to the basic model; An abnormal case database is established, the cases with an identification accuracy less than 85% or a fragmentation time more than 2 times the average value are separately labeled, the features are supplemented after expert labeling, and the model is retrained, and the database is evaluated for feature importance once every quarter, and the redundant features with a contribution degree less than 5% are eliminated.

[0024] Compared with the existing technology, the beneficial effects of the present application are:

[0025] In terms of stone composition identification, an innovative multimodal data acquisition and fusion technology is adopted to simultaneously acquire near-infrared spectroscopy and high-frequency ultrasound data. Combined with refined preprocessing and an improved deep residual network model, the accuracy and real-time performance of composition identification are significantly improved. Compared with single-modal detection, multimodal data can complement each other to overcome environmental interference and data incompleteness. The lightweight design and attention mechanism of the model ensure that the identification process is fast and focuses on key features, without relying on ex vivo samples or long-term detection. This meets the needs of real-time decision-making during surgery, completely eliminates the dependence on the doctor's subjective experience, and avoids treatment deviations caused by misjudgment of composition.

[0026] The intelligence and adaptability of lithotripsy parameter matching have been significantly improved. By establishing a composition-hardness correlation model, parameters are automatically corrected based on the stone's composition, eliminating the need for manual setting of fixed parameters. For stones with different compositions and varying physical properties, the system can dynamically adjust the shock wave energy, impact frequency, and focusing depth, ensuring effective fragmentation of hard stones while avoiding excessive impact on soft stones, thus significantly improving lithotripsy efficiency. Simultaneously, a real-time intraoperative feedback mechanism continuously monitors the stone fragmentation status, dynamically optimizing parameters based on fragment size and distribution density. This avoids incomplete fragmentation or tissue damage caused by fixed parameters, maximizing the protection of surrounding mucosal tissue, reducing the risk of complications such as hematuria and edema, and improving surgical safety.

[0027] This invention also possesses excellent clinical adaptability and continuous optimization capabilities. The preoperative multi-center database covers stone samples from different regions and of different types, and combined with auxiliary information such as metabolic indicators, it can adapt to diverse clinical cases. The postoperative database update mechanism adopts incremental learning and federated learning, which not only protects patient privacy but also continuously optimizes the model and parameter matching strategy through the accumulation of clinical data, enabling the system to continuously improve its accuracy and adaptability with clinical application. In addition, the coaxial integrated design of the probe, the water-filled balloon coupling device, and the dynamic tracking algorithm further enhance the convenience and stability of intraoperative operations, reduce the operational difficulty for medical staff, provide strong support for clinical promotion, and have significant application value in the field of minimally invasive treatment of urinary stones. Attached Figure Description

[0028] Fig. 1 This is a schematic block diagram of the method for rapid intraoperative identification of urinary stone components and adaptive matching of lithotripsy parameters proposed in this invention;

[0029] Fig. 2 This is a schematic diagram comparing the recognition accuracy of different stone components using the intraoperative rapid identification and lithotripsy parameter adaptive matching method for urinary stone components proposed in this invention.

[0030] Fig. 3 This is a schematic diagram illustrating the trend of lithotripsy time as a function of stone diameter in the method for rapid intraoperative identification of urinary stone components and adaptive matching of lithotripsy parameters proposed in this invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0033] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0034] Reference Figs. 1 to 3 A method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters, comprising the following steps:

[0035] Preoperatively constructing a stone composition-lithotripsy parameter database: collecting spectral feature libraries and ultrasound echo feature libraries of common stone compositions. The spectral feature library covers the characteristic absorption peaks of calcium oxalate, calcium phosphate, uric acid, cystine, and mixed stones in the 350-2500nm wavelength range. The ultrasound echo feature library contains the acoustic impedance, attenuation coefficient, and backscattering coefficient of stones with different compositions at frequencies of 5-15MHz. At the same time, it stores the optimal combination of lithotripsy parameters for the corresponding composition, including shock wave energy of 1-5J, shock frequency of 5-20Hz, and focusing depth of 5-15mm.

[0036] Intraoperative multimodal data real-time acquisition: Stone data is acquired synchronously by a miniature near-infrared spectral probe integrated into the lithotripsy device and a high-frequency ultrasound probe. The spectral probe sampling interval is 0.5nm, and the single acquisition time is ≤0.3s. The ultrasound probe adopts pulse-echo mode with a pulse repetition frequency of 1000Hz and a data sampling rate of 10MHz. The acquisition area covers the exposed surface of the stone and has a diameter ≥2mm to ensure that the data contains complete composition information.

[0037] Data preprocessing and feature extraction: Baseline correction, smoothing and denoising, and normalization were performed on the spectral data. SG filtering was used to remove high-frequency noise, with a window width of 11 points. The absorption peak features were enhanced by first-order derivative transformation, and absorbance values ​​at 5-8 characteristic wavelengths were extracted. Envelope detection and Fourier transform were performed on the ultrasonic data to extract the peak intensity, main peak frequency, and bandwidth features of the echo signal, forming a multi-dimensional feature vector.

[0038] Rapid identification of stone components: The extracted feature vectors are input into an improved deep residual network model. The model contains 4 residual blocks, each consisting of 2 convolutional layers and skip connections. The probability distribution of various stone components is output through the Softmax function. When the highest probability is ≥90%, the component is directly determined. When the probability is 80%-90%, clinical history is combined to assist in the determination. The determination time is ≤1s and the component identification accuracy is ≥95%.

[0039] Adaptive matching of lithotripsy parameters: Based on the recognition results, the initial parameter combination in the database is called, and the parameters are corrected by the established composition-hardness correlation model. When the hardness of calcium oxalate stones is high, the initial energy is increased by 10%-20%, and when the hardness of uric acid stones is low, the energy is reduced by 15%-25%. At the same time, the focusing depth is dynamically adjusted according to the stone diameter. For every 1 mm increase in diameter, the focusing depth increases by 0.5 mm.

[0040] Real-time feedback and parameter optimization: During the operation, the stone fragmentation status is monitored in real time by an ultrasound probe. Images of the fragmentation area are acquired every 0.5 seconds to analyze the average particle size and distribution density of the fragments. When the fragment size is ≤2mm and accounts for ≥80%, the impact frequency is reduced to 5-10Hz. When there is no significant change in the fragments after 3 consecutive impacts, the energy is increased by 5%-10% until the stone fragmentation reaches the target.

[0041] Postoperative parameter recording and database update: The system automatically records the stone composition, actual lithotripsy parameters used, and fragmentation effect evaluation of this surgery. The database is updated using an incremental learning algorithm, and the model is fine-tuned every 50 clinical cases to improve the accuracy of subsequent identification and parameter matching.

[0042] This invention also includes:

[0043] The confidence enhancement step in stone composition identification involves calculating the matching degree of the characteristic wavelength range through integration, using the following formula: Where C is the feature interval matching degree, λ is the wavelength, λ1 and λ2 are the start and end wavelengths of the feature interval, I(λ) is the spectral intensity of the stone to be identified, and W(λ) is the weighting function of the standard spectrum. The weighting function is 1.2-1.5 at the feature absorption peak position and 0.5-0.8 in the non-feature region. This matching degree is fused with the probability value output by the deep residual network at a weight of 6:4 to form the final judgment confidence. When the fused confidence is ≥85%, the component is confirmed. If it is lower, a second acquisition is initiated. Multi-dimensional verification improves the reliability of identifying mixed component stones and micro stones, which is especially suitable for mixed stones with similar component proportions.

[0044] The refined control steps for dynamic adjustment of lithotripsy energy are as follows: Energy output is corrected in real time based on the lithotripsy rate; the energy change rate formula is... in Let k be the rate of change of energy over time, and k be a proportionality constant, ranging from 0.02 to 0.05 and decreasing with increasing tissue distance. When the distance from the stone to the surrounding tissue is ≤2 mm, k is taken as 0.02. The volume change rate of stone fragmentation per unit time is calculated using ultrasound image sequences. H is the stone hardness value, estimated using a composition-hardness model. This dynamic adjustment causes the energy output to change stepwise with the fragmentation process. The energy increase rate is faster in the initial stage, and the rate decreases by 30% when the fragmentation rate reaches 50%, reducing the impact damage to surrounding tissues and shortening the fragmentation time.

[0045] In this invention, when constructing the stone composition-lithotomy parameter database before surgery, a stratified sampling combined with multi-center data fusion method is adopted. The samples cover patient groups from different regions, ensuring that the sample ratio of calcium oxalate, calcium phosphate, uric acid, cystine, and mixed stones is 4:2:2:1:1, with ≥300 samples for each component, including stones with diameters of 2-20mm and hardness values ​​of 30-150HV. The CT value of the stones, the pH value of the patient's urine, and metabolic indicators are recorded simultaneously. Stone samples from patients with metabolic abnormalities are marked as a special subset, and a parameter matching sub-model is trained separately for this subset. The database adopts a distributed storage architecture, supports real-time uploading of case data from multiple medical institutions, and performs cross-center data consistency verification once a month. Data offset caused by equipment differences is eliminated through Z-score standardization.

[0046] In this invention, during real-time acquisition of multimodal data during surgery, the spectral probe and ultrasound probe are coaxially integrated. The probe tip is equipped with a retractable water-filled balloon coupling device. The balloon is filled with sterile saline at 37°C, and the pressure inside the balloon is controlled at 5-10 mmHg by a pressure sensor to ensure close contact with the stone surface and avoid tissue compression. Spectral acquisition uses a ring-shaped LED array light source with a power of 50-100mW, which can be automatically adjusted according to the reflectivity of the stone surface. When the reflectivity is >60%, the power is reduced by 20%. The ultrasound probe uses phased array technology with a focusing area diameter ≤1mm, which can be adjusted in real time by electronic focusing. During acquisition, a combination of "spiral scanning + fixed-point sampling" mode is used. The spiral scanning covers the entire stone, and the fixed-point sampling is repeated 3 times for highly reflective areas to improve data integrity.

[0047] In this invention, during data preprocessing and feature extraction, the normalization of spectral data employs a standard normal variable transformation, as shown in the formula: in The normalized spectral intensity is represented by I(λ), the original spectral intensity is represented by μ, the mean spectral intensity is represented by σ, and the standard deviation is represented by σ. After processing, the validity is verified by calculating the cosine similarity with the standard sample. A similarity ≥ 0.95 is considered qualified; otherwise, the abnormal data repair process is initiated. The Fourier transform of the ultrasound data is performed using short-time Fourier transform with a time window width of 512 points and an overlap rate of 75%. The power spectral density of 3-5 feature frequency points is extracted, and the energy proportion of 5 frequency bands is obtained by combining wavelet packet decomposition to form a 10-dimensional ultrasound feature vector, which enhances the differentiation of stones with different compositions.

[0048] In this invention, for rapid identification of stone components, an improved deep residual network model introduces a hybrid attention mechanism. Channel attention layers and spatial attention layers are added sequentially after each residual block. The channel attention layer learns the weights of spectral bands and ultrasound features through a squeeze-excitation operation, while the spatial attention layer focuses on the feature differences between the stone's edge and core regions. The model employs a lightweight design, reducing the number of convolutional kernels by 30% and introducing depthwise separable convolutions, thus reducing computation by 40% and meeting the real-time requirements during surgery. Model training uses a hybrid loss function, composed of cross-entropy loss and center loss weighted in an 8:2 ratio. The center loss constrains the clustering of features of similar components. The batch size is set to 32, the number of iterations is 500, the initial learning rate is 0.001, and it decays by 10% every 100 iterations. An early stopping strategy prevents overfitting, and training terminates when the validation set accuracy shows no improvement after 50 consecutive iterations.

[0049] In this invention, during the adaptive matching of crushed stone parameters, the composition-hardness correlation model is established through multivariate nonlinear regression. The relationship between the hardness value H and the characteristic wavelength absorbance A1 (980nm), A2 (1450nm), ultrasonic acoustic impedance Z, and attenuation coefficient α is as follows: Where a, b, c, and d are regression coefficients, and e is a constant term, obtained by fitting data from 500 clinical cases. The coefficients are a=12, b=18, c=5, d=0.8, and e=3 for calcium oxalate stones, a=10, b=15, c=4, d=0.6, and e=2 for calcium phosphate stones, and a=8, b=12, c=3, d=0.5, and e=1 for uric acid stones. A stone location correction factor was also introduced during parameter matching, which reduced the impact frequency of stones located in the renal pelvis by 10% and improved the focusing depth accuracy of stones located in the upper ureter to ±0.2mm, ensuring that the parameters are adapted to the differences in anatomical location.

[0050] In this invention, during real-time feedback and parameter optimization, the average particle size of the fragments is calculated using a multi-scale feature fusion method based on ultrasound images. Gray-level co-occurrence matrix features and LBP features of 10×10, 20×20, and 30×30 pixel windows are extracted and input into the particle size estimation model to obtain the fragment size, with a model error ≤0.3mm. When the stone displacement exceeds 1mm, a dynamic tracking algorithm is activated. The displacement vector is calculated using five consecutive frames of images to predict the trajectory and adjust the focusing depth 50ms in advance. The trajectory prediction uses second-order polynomial fitting. A tissue protection threshold is set. When the distance between the shock wave focus and the mucosa is ≤1.5mm, the energy is automatically reduced by 30% and the impact interval is increased to 0.5s to avoid tissue damage. Simultaneously, the tissue elasticity characteristics of the ultrasound echo are used to monitor the damage risk. When the elastic coefficient change rate is >15%, the impact is paused and a warning is issued.

[0051] In this invention, when recording surgical parameters and updating the database after surgery, the fragmentation effect is evaluated using the fragmentation efficiency index E, defined as the ratio of the total mass of fragments with a diameter ≤2mm after fragmentation to the mass of the original stone. Combined with the surgical time T and patient blood loss B, a comprehensive score S is calculated as: S = 0.5·E + 0.3·(1 / T) + 0.2·(1 / B). A higher score indicates a better effect. Incremental learning employs a federated learning framework, where each medical institution trains the model locally, uploading only the updated model parameters to protect patient privacy. Every 50 data points are used for model fine-tuning, employing knowledge distillation to transfer knowledge from new data to the basic model. An abnormal case database is established, with cases having an accuracy rate <85% or fragmentation time exceeding twice the average separately labeled. After supplementing features through expert annotation, the model is retrained to improve its adaptability to complex cases. The database undergoes a feature importance assessment quarterly, eliminating redundant features with a contribution of <5%.

[0052] Specific implementation method of rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters:

[0053] Example 1

[0054] Implementation of Extracorporeal Shock Wave Lithotripsy for Single Calcium Oxalate Stones

[0055] This example describes the intraoperative management of a single-component urinary calculus, namely calcium oxalate calculi. The patient was a 45-year-old male, clinically diagnosed with renal pelvis calcium oxalate calculi. The calculi were 10 mm in diameter, with a hardness of approximately 80 HV. The patient had no history of metabolic abnormalities. The preoperative CT value was 1200 HU, and the urine pH was 6.5. The procedure strictly followed the entire technical protocol, ensuring accurate component identification and appropriate lithotripsy parameters through refined parameter settings and operational procedures. The specific steps are as follows:

[0056] 1. Preoperative database construction and preparation

[0057] Preoperatively, a stone composition-lithotripsy parameter database was accessed. This database, constructed through multi-center data fusion, covers clinical samples from six regions across China, including 420 cases of calcium oxalate stones. These stones ranged in diameter from 2-20 mm and hardness from 30-150 HV, and auxiliary information such as patient CT values ​​and urine pH were recorded simultaneously. For calcium oxalate stones, the database stored characteristic spectra with core absorption peaks at 980 nm and 1450 nm, and the ultrasound characteristic parameter was an acoustic impedance of 3.5 × 10⁻⁶. 6 kg / (m 2The optimal initial combination of lithotripsy parameters is: shock wave energy 3.5J, shock frequency 15Hz, and focusing depth 10mm. The database employs distributed storage, and cross-center data consistency verification is completed 24 hours preoperatively. Z-score standardization eliminates data offsets between different devices, ensuring the reliability of the retrieved parameters.

[0058] 2. Real-time acquisition of intraoperative multimodal data

[0059] An integrated miniature near-infrared spectral probe and a high-frequency ultrasonic probe are used. The probes are coaxial and equipped with a retractable water-filled coupling device at the front end. The water-filled balloon is filled with sterile saline at 37°C, and the pressure inside the balloon is stabilized at 8 mmHg by a pressure sensor to ensure close contact with the stone surface without tissue compression. Spectral acquisition uses a ring-shaped LED array light source with an initial power setting of 80mW. The reflectivity of the detected stone surface is 45%, and this value is less than 60%, so no power adjustment is required. The spectral probe sampling interval is 0.5nm, the single acquisition time is 0.25s, and three consecutive acquisitions are taken and the average value is taken. The acquisition area diameter is 2.5mm, covering the core area of ​​the exposed stone surface. The ultrasonic probe uses phased array technology. In pulse-echo mode, the pulse repetition frequency is 1000Hz, the data sampling rate is 10MHz, and the focusing area diameter is 0.8mm. It adopts a "spiral scan + fixed-point sampling" mode: the spiral scan speed is 5mm / s, covering the entire stone; for the highly reflective area at the edge of the stone with a reflectivity of 55%, fixed-point repeated acquisition is performed three times to ensure data integrity.

[0060] 3. Data Preprocessing and Feature Extraction

[0061] The spectral data underwent preprocessing as follows: First, SG filtering was used for noise reduction with an 11-point window to eliminate high-frequency interference; then, first-derivative transformation was applied to enhance the absorption peak characteristics, highlighting the absorbance variations at 980 nm and 1450 nm; finally, standard normalization was performed, using the formula: The original spectrum has an intensity of I(λ) = 0.72 at 980 nm, a mean intensity μ = 0.68, and a standard deviation σ = 0.12. The normalized intensity is then calculated. At 1450 nm, I(λ) = 0.58, calculated as follows: After normalization, the cosine similarity with the standard calcium oxalate sample was calculated to be 0.97, and this value was greater than or equal to 0.95, thus the data was deemed acceptable. Ultrasonic data preprocessing: First, envelope detection was performed to extract the echo signal profile. Then, short-time Fourier transform was used to analyze the frequency characteristics. The time window width was 512 points, the overlap rate was 75%, and the power spectral density at three characteristic frequency points of 2MHz, 5MHz, and 8MHz was extracted, which were 0.6×10⁻⁶ respectively. -6 W / Hz, 1.2×10 -6 W / Hz, 0.8×10- 6 W / Hz; combined with wavelet packet decomposition, the energy percentages of five frequency bands, namely 0-2MHz, 2-4MHz, 4-6MHz, 6-8MHz, and 8-10MHz, are obtained, which are 15%, 25%, 30%, 20%, and 10% respectively, forming a 10-dimensional ultrasonic feature vector. This is combined with the spectral features, namely the absorbance at 980nm and 1450nm, to form a 12-dimensional total feature vector.

[0062] 4. Stone composition identification and lithotripsy parameter matching

[0063] A 12-dimensional feature vector is input into an improved deep residual network model. The model contains four residual blocks, each followed by a channel attention layer and a spatial attention layer. The channel attention layer assigns a weight of 0.6 to spectral features and 0.4 to ultrasound features; the spatial attention layer assigns a weight of 0.7 to the core region of the stone and 0.3 to the peripheral region. The model has a batch size of 32, 500 iterations, and a current learning rate of 0.0001 (this value is the result after 400 iterations of decay). The component probability distribution is output using the Softmax function, showing a 96% probability for calcium oxalate stones and probabilities of less than 5% for other components.

[0064] Simultaneously calculate the matching degree of the characteristic wavelength range, using the following formula: The characteristic ranges for calcium oxalate were selected as λ1 = 970 nm and λ2 = 990 nm. The average spectral intensity of the stone to be identified in this range was I(λ) = 0.7. The standard spectral weighting function W(λ) = 1.4, where this value represents the weight corresponding to the characteristic peak position. The integral calculation yielded C = 0.7 × 1.4 × (990 - 970) × 10. -9 ×10 3 After adding the unit conversion factor, the result is approximately 1.96 × 10⁻⁶. -4 The matching score C and the model output probability are fused with a weighted average of 6:4, resulting in a final confidence score of 0.6 × (1.96 × 10⁻⁶). -4 / 2×10 -4 ) + 0.4 × 0.96 ≈ 0.6 × 0.98 + 0.4 × 0.96 ≈ 0.972, which is 97.2%, confirming that the stone is calcium oxalate.

[0065] Based on the identification results, initial parameters are retrieved from the database, and then the parameters are corrected using a composition-hardness correlation model. The model formula is as follows: Where A1 is the absorbance at 980nm = 0.6, A2 is the absorbance at 1450nm = 0.4, and Z is the acoustic impedance = 3.5 × 10⁻⁶. 6 kg / (m 2·s), α is the attenuation coefficient = 0.8 dB / mm, and the corresponding coefficients for calcium oxalate are a = 12, b = 18, c = 5, d = 0.8, e = 3. Therefore, H = 12 × 0.6. 2 +18×0.4+5×ln(3.5×10 6 The formula is: 0.8 × 0.8 + 3 ≈ 12 × 0.36 + 7.2 + 5 × 15.07 + 0.64 + 3 ≈ 4.32 + 7.2 + 75.35 + 0.64 + 3 ≈ 90.51 HV. This value is close to the actual hardness of 80 HV, and the error is within the allowable range. Because the hardness is slightly higher than the average level, the initial energy is increased by 15% from 3.5 J to 4.0 J, the focusing depth is kept at 10 mm, and this value is adapted to the stone diameter of 10 mm. The impact frequency is maintained at 15 Hz.

[0066] 5. Real-time feedback and parameter optimization

[0067] During the procedure, images of the lithotripsy area were acquired every 0.5 seconds using an ultrasound probe. The average fragment size was calculated using a multi-scale feature fusion method: the gray-level co-occurrence matrix contrasts of 10×10, 20×20, and 30×30 pixel windows were extracted, with LBP feature entropy values ​​of 1.2, 1.0, and 0.8 respectively. These values ​​were input into the fragment size estimation model, with an initial fragment size of 4 mm, accounting for 30%. The energy was adjusted according to the fragmentation rate, using the following formula: [Energy Change Rate Formula]. Where k = 0.03, and this value corresponds to a distance of 3 mm between the stone and the surrounding tissue, and this distance is greater than 2 mm. dV / dt, i.e., the fragmentation volume per unit time, = 2 mm. 3 The value was calculated from the image sequence, H = 90.51HV, and dE / dt = 0.03 × 2 × 90.51 ≈ 5.43 J / s. Initially, the energy was increased at this rate, reaching 4.2 J after 5 seconds. The proportion of fragments with a diameter ≤ 2 mm reached 50%. At this point, the energy change rate was reduced by 30%, and dE / dt was adjusted to 3.8 J / s.

[0068] When there was no significant change in fragment size after three consecutive 1.5s monitoring sessions, and the fragment size remained at 2mm (55% of the total), the energy was increased by 8% to 4.5J, while the frequency was maintained at 15Hz. After 10s, when the fragment size ≤2mm accounted for 85%, the frequency was reduced to 8Hz, while the energy was maintained at 4.5J. During this period, a stone displacement of 0.8mm was detected, and this value was less than 1mm, so no adjustment of the focusing depth was required; the distance between the shock wave focus and the mucosa was 2mm, and this value was greater than 1.5mm, so no reduction in energy was required. The final lithotripsy time was 15 minutes, and the fragmentation rate was 90%.

[0069] 6. Post-operative record and database update

[0070] The data for this surgery was automatically recorded: the stone composition was calcium oxalate; actual lithotripsy parameters included energy 4.0-4.5J, frequency 8-15Hz, depth of focus 10mm, fragmentation effect including fragmentation efficiency index E=0.9, operation time T=15 minutes, blood loss B=3ml, and comprehensive score S=0.5×0.9+0.3×(1 / 15)+0.2×(1 / 3)≈0.45+0.02+0.067≈0.537. A federated learning framework was used; after local data training, only the model parameter updates were uploaded, with weight adjustments less than 5%, protecting patient privacy. This data was included in the database, and the model will be fine-tuned after accumulating 50 cases.

[0071] 7. Effect Verification and Table Analysis

[0072] Table 1 Key indicators for calcium oxalate stone treatment in Example 1:

[0073]

[0074] Table 1 shows the data based on this surgical procedure and a one-month follow-up monitoring, visually demonstrating the technical effectiveness of this invention. The component identification accuracy reached 97.2%, thanks to multimodal data fusion and confidence enhancement. Even single-component stones can be accurately identified through complementary verification using spectroscopy and ultrasound. The lithotripsy time was 15 minutes, with a fragmentation rate of 90%, proving the effectiveness of adaptive parameter matching. The energy output based on hardness adjustment can quickly fragment stones while avoiding excessive impact. The complication rate was 0, reflecting the tissue protection effect of real-time feedback. By monitoring the distance between the focal point and the mucosa and stone displacement, the risk of damage is minimized. After model updates, the accuracy improved to 98.5%, validating the value of incremental database learning. With the accumulation of clinical data, the system's identification capabilities continue to optimize, adapting to more cases of single-component stones.

[0075] Example 2

[0076] Implementation of Extracorporeal Shock Wave Lithotripsy for Mixed-Component Stones

[0077] This example describes the intraoperative management of mixed secretory urinary stones, specifically a mixture of calcium oxalate and calcium phosphate in a 6:4 ratio. The patient was a 52-year-old female, clinically diagnosed with mixed stones in the upper ureter, with a stone diameter of 12 mm and a hardness of approximately 70 HV. The patient had a history of hyperuricemia and metabolic abnormalities. Preoperative CT value was 1000 HU, and urine pH was 5.8. The procedure focused on identifying the mixed components and dynamically adjusting multiple parameters. The specific steps are as follows:

[0078] 1. Preoperative database and special subset access

[0079] Preoperatively, a stone composition-lithotripsy parameter database was used. This database contained 350 mixed stone samples, including 140 cases of calcium oxalate-calcium phosphate mixed stones. Simultaneously, samples from patients with metabolic abnormalities, including those with hyperuricemia, were designated as a special subset, and the weights of the sub-model trained for this subset were adjusted by 10%. The spectral characteristics of mixed stones were a double absorption peak at 980 nm (calcium oxalate) and 1100 nm (calcium phosphate). The ultrasound characteristics were an acoustic impedance of 3.2 × 10⁻⁶. 6 kg / (m 2 •s), attenuation coefficient 0.7dB / mm; the initial lithotripsy parameter combination was energy 3J, frequency 12Hz, and focusing depth 10.5mm. The calculation method was to add (12-8)×0.5mm to the base depth corresponding to a stone diameter of 12mm, resulting in 10mm. Due to the mixed components, it was slightly adjusted to 10.5mm. Data consistency verification was completed preoperatively, and the parameters of the metabolic abnormality sub-model were loaded.

[0080] 2. Intraoperative multimodal data acquisition

[0081] Using the same coaxial integrated probe as in Example 1, the pressure inside the balloon was adjusted to 7 mmHg, taking into account the fragility of the upper ureteral tissue. The initial power of the spectral light source was 70 mW, and the reflectivity of the stone surface was detected at 65%, which was greater than 60%. The power was then reduced by 20% to 56 mW. The spectral sampling interval was 0.5 nm, the single acquisition time was 0.3 s, and four consecutive acquisitions were taken, with the average value taken. The acquisition area diameter was 3 mm, covering the different component distribution areas of the mixed stone. The ultrasound probe focusing area diameter was 1 mm, the pulse repetition frequency was 1000 Hz, and a spiral scan was used at a speed of 4 mm / s with dual fixed-point sampling: 60% reflectivity in the calcium oxalate area and 50% reflectivity in the calcium phosphate area. Three acquisitions were taken at each fixed point to ensure that data for both components were captured.

[0082] 3. Data Preprocessing and Feature Extraction

[0083] Spectral data preprocessing: After 11 points of SG filtering, the first derivative transformation highlights the double absorption peaks at 980nm and 1100nm; standard normalization transformation is performed, formula... At 980 nm, I(λ) = 0.65, μ = 0.62, σ = 0.11, and the calculated values ​​are... At 1100 nm, I(λ) = 0.55, calculated as follows: The cosine similarity with the standard mixed sample is 0.96, which is acceptable.

[0084] Ultrasonic data preprocessing: Short-time Fourier transform window of 512 points, overlap rate of 75%, extracting power spectral densities of 2MHz, 5MHz, and 8MHz, respectively, which are 0.5×10⁻⁶. -6 W / Hz, 1.0×10 -6W / Hz, 0.7×10 -6 W / Hz; wavelet packet decomposition yields 5 frequency bands with energy percentages of 18%, 22%, 28%, 20%, and 12%, forming a 10-dimensional ultrasonic feature vector, which is combined with the absorbance of the double absorption peaks to form a 12-dimensional total feature vector.

[0085] 4. Component identification and parameter matching

[0086] The feature vector is input into the deep residual network. The model attention layer assigns a weight of 0.35 to 980nm, a weight of 0.3 to 1100nm, a weight of 0.2 to ultrasonic impedance, and a weight of 0.15 to the attenuation coefficient. The softmax output probabilities are: calcium oxalate 62%, calcium phosphate 35%, and others 3%, with the highest probability being 62% and this value being less than 90%.

[0087] The formula for calculating the matching degree of two feature intervals is: For calcium oxalate, the wavelength ranges are λ1 = 970 nm and λ2 = 990 nm. Given I(λ) = 0.65 and W(λ) = 1.3, integrating yields C1 = 0.65 × 1.3 × 20 × 10⁻⁶. -9 ×10 3 ≈1.69×10 -4 For calcium phosphate, the wavelength ranges are λ1 = 1090 nm and λ2 = 1110 nm. Given I(λ) = 0.55 and W(λ) = 1.2, integrating yields C2 = 0.55 × 1.2 × 20 × 10⁻⁶. -9 ×10 3 ≈1.32×10 -4 Fusion confidence level = 0.6 × [(1.69 × 10⁻⁶)] -4 +1.32×10 -4 ) / (2×10 -4 +1.8×10 -4 )]+0.4×(0.62+0.35)≈0.6×(3.01×10 -4 / 3.8×10 -4 )+0.4×0.97≈0.6×0.792+0.388≈0.863, which is 86.3%. Considering the patient's history of hyperuricemia and the fact that this information increases the probability of mixed components by 5%, it was finally confirmed that calcium oxalate and calcium phosphate were mixed with a ratio of 6:4.

[0088] Hardness is estimated using a composition-hardness model. The formula uses the average of calcium oxalate and calcium phosphate as the mixing composition coefficient. b=(18+15) / 2=16.5, c=(5+4) / 2=4.5, d=(0.8+0.6) / 2=0.7, e=(3+2) / 2=2.5; A1=0.65, A2=0.55, Z=3.2×10 6 kg / (m 2Given α = 0.7 dB / mm, we calculate H = 11 × 0.65 · s), α = 0.7 dB / mm. 2 +16.5×0.55+4.5×ln(3.2×10 6 The formula is: 4.5 × 15.08 + 0.49 + 2.5 ≈ 11 × 0.4225 + 9.075 + 4.5 × 15.08 + 0.49 + 2.5 ≈ 4.6475 + 9.075 + 67.86 + 0.49 + 2.5 ≈ 84.57 HV. This value is close to the actual 70 HV, and the error is acceptable. Because the mixed components have moderate hardness, the energy is taken as 3.2 J, which is the average of the initial 3.5 J for calcium oxalate and the initial 2.8 J for calcium phosphate. The frequency is 12 Hz, and the focusing depth is 10.5 mm, corresponding to a stone with a diameter of 12 mm. A position correction factor is introduced, specifically for the upper ureter, improving the focusing depth accuracy to ±0.2 mm.

[0089] 5. Real-time feedback and parameter optimization

[0090] Ultrasonic images are acquired every 0.5 seconds. The initial particle size of the debris is 5 mm, accounting for 20%. Energy change rate formula. k = 0.025, and this value corresponds to a distance of 2.5 mm between the stone and the mucosa, with dV / dt = 1.8 mm. 3 / s, H=84.57HV, calculated dE / dt=0.025×1.8×84.57≈3.805J / s. Initial energy 3.2J, rising to 3.4J after 5s, fragments ≤2mm accounted for 40%; after 10s, stone displacement of 1.2mm was detected, and this value was greater than 1mm, dynamic tracking was initiated, the displacement vector was calculated through 5 frames of images, i.e. X direction +0.8mm, Y direction +0.4mm, second-order polynomial fitting predicted trajectory, and the focusing depth was adjusted to 10.7mm 50ms in advance.

[0091] When 70% of the fragments were ≤2mm, the distance between the shock wave focal point and the mucosa was 1.5mm, which was within the set threshold. The energy was automatically reduced by 30% to 2.38J, and the impact interval increased to 0.5s. After three consecutive impacts, the fragment percentage remained unchanged, and the energy increased by 7% to 2.55J. The final lithotripsy time was 20 minutes, with a fragmentation rate of 85%. During the process, the tissue elasticity coefficient changed by 12%, and this value was less than 15%, with no warning triggered.

[0092] 6. Post-operative record and database update

[0093] Recorded data: The mixed composition is calcium oxalate and calcium phosphate in a 6:4 ratio; parameters include energy 2.38-3.4J, frequency 12Hz, depth of focus 10.5-10.7mm; fragmentation effect includes E=0.85, T=20 minutes, B=5ml; and the overall score S=0.5×0.85+0.3×(1 / 20)+0.2×(1 / 5)=0.425+0.015+0.04=0.48. The data is labeled as "metabolic abnormality - mixed stones" and included in a special subset. Federated learning is used to upload parameter updates. After this data accumulation, one new case was added to the abnormal case database, and the recognition accuracy of this case was 86.3%, less than 90%. Subsequent retraining will be carried out with additional features through expert annotation.

[0094] 7. Effect Verification and Table Analysis

[0095] Table 2 Key Indicators for Mixed-Component Stone Treatment in Example 2:

[0096]

[0097] Table 2 shows data based on surgery and postoperative follow-up, demonstrating the adaptability of this invention to mixed-component stones. The component identification accuracy rate is 86.3%, lower than that of single-component stones, but through confidence fusion and medical history assistance, it can still accurately determine the mixing ratio, overcoming the bottleneck of traditional methods' difficulty in distinguishing mixed components. The lithotripsy time is 20 minutes, and the fragmentation rate is 85%. Due to the large differences in the physical properties of the mixed components, parameters need to be dynamically balanced to meet the needs of both components, yet efficient fragmentation is still achieved. The complication rate is 1%, i.e., mild hematuria, proving the effectiveness of the tissue protection mechanism; even in the vulnerable area of ​​the upper ureter, the risk can be reduced through energy adjustment and displacement tracking. The special subset adaptation rate is 92%, verifying the database's targeting of patients with metabolic abnormalities. Sub-model training improves the ability to handle complex cases, providing a reliable solution for the clinical treatment of mixed-component stones.

[0098] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters, characterized in that, include: Preoperatively constructing a stone composition-lithotripsy parameter database: collecting spectral feature libraries and ultrasound echo feature libraries of common stone compositions. The spectral feature library covers the characteristic absorption peaks of calcium oxalate, calcium phosphate, uric acid, cystine, and mixed stones in the 350-2500nm wavelength range. The ultrasound echo feature library contains the acoustic impedance, attenuation coefficient, and backscattering coefficient of stones with different compositions at frequencies of 5-15MHz. At the same time, it stores the optimal combination of lithotripsy parameters for the corresponding composition, including shock wave energy, shock frequency, and focusing depth. Intraoperative real-time data acquisition: Stone data is acquired synchronously by a miniature near-infrared spectral probe integrated into the lithotripsy device and a high-frequency ultrasonic probe. The ultrasonic probe adopts a pulse-echo mode, and the acquisition area covers the exposed surface of the stone with a diameter ≥2mm. Data preprocessing and feature extraction: Baseline correction, smoothing and denoising, and normalization were performed on the spectral data. SG filtering was used to remove high-frequency noise. The window width was set to 11 points. The absorption peak features were enhanced by first-order derivative transformation, and absorbance values ​​at 5-8 characteristic wavelengths were extracted. Envelope detection and Fourier transform were performed on the ultrasonic data to extract the peak intensity, main peak frequency, and bandwidth features of the echo signal to form a feature vector. Stone composition identification: The extracted feature vector is input into a deep residual network model. The model contains 4 residual blocks, each consisting of 2 convolutional layers and skip connections. The probability distribution of various stone components is output through the Softmax function. When the highest probability is ≥90%, the component is directly determined. When the probability is 80%-90%, clinical history is combined to assist in the determination. The determination time is ≤1s and the component identification accuracy is ≥95%. Adaptive matching of lithotripsy parameters: Based on the recognition results, the initial parameter combination in the database is called, and the parameters are corrected by the established composition-hardness correlation model. When the hardness of calcium oxalate stones is high, the initial energy is increased by 10%-20%, and when the hardness of uric acid stones is low, the energy is reduced by 15%-25%. At the same time, the focusing depth is dynamically adjusted according to the stone diameter. For every 1 mm increase in diameter, the focusing depth increases by 0.5 mm. Real-time feedback and parameter optimization: During the operation, the stone fragmentation status is monitored in real time by an ultrasound probe. Images of the fragmentation area are acquired every 0.5 seconds to analyze the average particle size and distribution density of the fragments. When the fragment size is ≤2mm and accounts for ≥80%, the impact frequency is reduced to 5-10Hz. When there is no significant change in the fragments after 3 consecutive impacts, the energy is increased by 5%-10% until the stone fragmentation reaches the target. Postoperative parameter recording and database update: The system automatically records the stone composition, actual lithotripsy parameters used, and fragmentation effect evaluation of this surgery. The database is updated using an incremental learning algorithm, and the model is fine-tuned every 50 clinical cases.

2. The method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters according to claim 1, characterized in that, Also includes: The confidence enhancement step in stone composition identification involves calculating the matching degree of the characteristic wavelength range through integration, using the following formula: Where C is the feature interval matching degree, λ is the wavelength, λ1 and λ2 are the start and end wavelengths of the feature interval, I(λ) is the spectral intensity of the stone to be identified, and W(λ) is the weighting function of the standard spectrum. When the fusion confidence is ≥85%, the component is confirmed; otherwise, a second acquisition is initiated. Verification improves the reliability of identifying mixed component stones and micro stones.

3. The method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters according to claim 1, characterized in that, Also includes: The refined control steps for dynamic adjustment of lithotripsy energy are as follows: Energy output is corrected in real time based on the lithotripsy rate; the energy change rate formula is... in Let k be the rate of change of energy over time, and k be a proportionality constant, ranging from 0.02 to 0.05 and decreasing with increasing tissue distance. When the distance from the stone to the surrounding tissue is ≤2 mm, k is taken as 0.

02. The volume change rate of stone fragmentation per unit time is calculated using ultrasound image sequences. H is the stone hardness value, estimated using a composition-hardness model. The energy output is dynamically adjusted to change stepwise with the fragmentation process. The energy increase rate is faster in the initial stage, and the rate decreases by 30% when the fragmentation rate reaches 50%, reducing the impact damage to surrounding tissues and shortening the fragmentation time.

4. The method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters according to claim 1, characterized in that, When constructing the preoperative stone composition-lithotripsy parameter database, a stratified sampling method combined with multi-center data fusion was adopted. The samples included patient groups from different regions. The CT value of the stones, the pH value of the patients' urine, and metabolic indicators were recorded simultaneously. Stone samples from patients with metabolic abnormalities were marked as a special subset. A parameter matching sub-model was trained separately for the subset. The database adopted a distributed storage architecture to support real-time uploading of case data by various medical institutions.

5. The method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters according to claim 1, characterized in that, During real-time acquisition of intraoperative data, the spectral probe and the ultrasound probe are coaxially integrated. The probe front end is equipped with a retractable water-filled balloon coupling device. The balloon is filled with sterile saline at 37°C. The pressure inside the balloon is controlled at 5-10 mmHg by a pressure sensor. The spectral acquisition uses a ring LED array light source with a power of 50-100mW, which is automatically adjusted according to the reflectivity of the stone surface.

6. The method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters according to claim 1, characterized in that, In the data preprocessing and feature extraction, the normalization of the spectral data adopts a standard normal variable transformation, as shown in the formula: in I(λ) represents the normalized spectral intensity, μ represents the mean spectral intensity, and σ represents the standard deviation. After processing, the validity is verified by calculating the cosine similarity with the standard sample. A similarity ≥ 0.95 is considered acceptable; otherwise, the abnormal data repair process is initiated.

7. The method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters according to claim 1, characterized in that, In the stone component identification process, the deep residual network model introduces a hybrid attention mechanism. After each residual block, a channel attention layer and a spatial attention layer are added sequentially. The channel attention layer learns the weights of spectral bands and ultrasound features through a squeeze-excitation operation, while the spatial attention layer focuses on the feature differences between the stone's edge and core regions. The model adopts a lightweight design. The model training uses a hybrid loss function, consisting of a cross-entropy loss and a center loss weighted in an 8:2 ratio. The center loss constrains the clustering of features of similar components. The batch size is set to 32, the number of iterations is 500, the initial learning rate is 0.001, and it decreases by 10% every 100 iterations. An early stopping strategy is used to prevent overfitting, and training is terminated when the accuracy on the validation set shows no improvement after 50 consecutive iterations.

8. The method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters according to claim 1, characterized in that, In the adaptive matching of crushed stone parameters, the composition-hardness correlation model is established through multivariate nonlinear regression. The relationship between the hardness value H and the characteristic wavelength absorbance A1, A2, ultrasonic impedance Z, and attenuation coefficient α is as follows: Where a, b, c, and d are regression coefficients, and e is a constant term, obtained by fitting data from 500 clinical cases. The coefficients are a=12, b=18, c=5, d=0.8, and e=3 for calcium oxalate stones, a=10, b=15, c=4, d=0.6, and e=2 for calcium phosphate stones, and a=8, b=12, c=3, d=0.5, and e=1 for uric acid stones. A stone location correction factor was also introduced during parameter matching, which reduced the impact frequency of stones located in the renal pelvis by 10% and improved the focusing depth accuracy of stones located in the upper ureter to ±0.2mm.

9. The method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters according to claim 1, characterized in that, In the real-time feedback and parameter optimization, the average particle size of the fragments is calculated using a multi-scale feature fusion method based on ultrasound images. The gray-level co-occurrence matrix features and LBP features of 10×10, 20×20, and 30×30 pixel windows are extracted and input into the particle size estimation model to obtain the fragment size. When the stone displacement exceeds 1 mm, the dynamic tracking algorithm is activated. The displacement vector is calculated through 5 consecutive frames of images to predict the motion trajectory and adjust the focus depth 50 ms in advance. The trajectory prediction uses second-order polynomial fitting. A tissue protection threshold is set. When the distance between the shock wave focus and the mucosa is ≤1.5 mm, the energy is automatically reduced by 30% and the impact interval is increased to 0.5 s to avoid tissue damage. At the same time, the tissue elasticity characteristics of the ultrasound echo are used to monitor the risk of damage. When the elastic coefficient change rate is >15%, the impact is paused and an early warning is issued.

10. The method for rapid intraoperative identification of urinary calculi components and adaptive matching of lithotripsy parameters according to claim 1, characterized in that, The evaluation of fragmentation effect during postoperative parameter recording and database updates uses the fragmentation efficiency index E, defined as the ratio of the total mass of fragments with a diameter ≤2mm after fragmentation to the original stone mass. Combined with the operation time T and patient blood loss B, a comprehensive score S is calculated as: S = 0.5·E + 0.3·(1 / T) + 0.2·(1 / B). A higher score indicates better performance. Incremental learning employs a federated learning framework, where each medical institution trains the model locally, uploading only the updated model parameters to protect patient privacy. Model fine-tuning is performed every 50 data points, using knowledge distillation to transfer knowledge from new data to the base model. An abnormal case database is established, with cases having an accuracy rate <85% or fragmentation time exceeding twice the average separately labeled. After expert annotation to supplement features, the database is retrained. Feature importance is assessed quarterly, eliminating redundant features with a contribution of <5%.

Images