Oral pulp cleaning and treatment method and pulp treatment system

Abstract

The present invention provides an oral pulp cleaning and treatment method and a pulp treatment system thereof, which relate to the technical field of the oral industry, including: fusing multimodal images such as CBCT and OCT to construct a four-dimensional risk model to guide path planning; performing adaptive and precise debridement under closed-loop feedback of multiple sensors such as pressure, acoustics, and chemistry; designing and 3D printing personalized restorations that are compatible with biomechanics through topological optimization algorithms; and finally achieving long-term stability monitoring and early failure warning of affected teeth through implantable passive sensors. The present invention also discloses an intelligent processing system for implementing this method. Through an integrated intelligent processing process, the present invention significantly improves the safety, success rate, and long-term survival rate of complex root canal treatments.

Description

Technical Field

[0001] The present invention relates to the technical field of oral industry, and in particular to an oral pulp cleaning and treatment method and a pulp treatment system thereof. Background Art

[0002] Root canal therapy is currently the main treatment method in the field of oral medicine to preserve teeth with pulp necrosis or infection due to pulp disease, periapical periodontitis, etc. Its core goal is to completely remove the infected substances in the root canal system, including necrotic pulp tissue, bacteria and their metabolites, and then perform strict three-dimensional filling and sealing of the root canal system to prevent reinfection and ultimately restore the chewing function of the affected tooth. Existing conventional root canal treatment technology still has many limitations in multiple key links such as diagnosis, debridement, restoration and prognosis monitoring, which directly affects the long-term success rate of treatment:

[0003] Current diagnosis relies primarily on two-dimensional periapical radiographs and three-dimensional cone-beam computed tomography (CBCT). Radiographs suffer from image overlap, and while CBCT can provide three-dimensional structures, its spatial resolution is insufficient to clearly display the microanatomical structures within the root canal system, such as the complex morphology of accessory canals, isthmuses, and C-shaped root canals. Furthermore, it is impossible to directly observe microcracks on the root canal walls or the actual attachment of biofilm. Furthermore, existing diagnostics rely solely on static assessments, unable to dynamically predict the future development and aggressiveness of the lesion, resulting in a lack of foresight in treatment planning.

[0004] After treatment is complete, long-term outcomes for the affected tooth are primarily assessed based on the presence of clinical symptoms and regular X-ray follow-up. This passive monitoring approach often indicates recurrence of infection or significant restoration failure by the time clinical symptoms appear or X-rays reveal a significant bone defect, potentially missing the optimal opportunity for remedial treatment. Summary of the Invention

[0005] Technical problems solved

[0006] In view of the deficiencies of the prior art, the present invention provides an oral pulp cleaning and treatment method and a pulp treatment system thereof, which solve the problems of the prior art.

[0007] Technical Solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for cleaning and treating oral pulp, comprising the following steps:

[0009] Sp1. Multimodal digital twin construction and risk prediction:

[0010] By fusing three-dimensional structural data of dental hard tissue obtained by cone-beam computed tomography with the microstructure and blood perfusion data of the soft tissue inside the dental pulp cavity obtained by optical coherence tomography, a multi-scale, high-fidelity digital twin model of the dentin-pulp complex was constructed. The imaging features of the lesion area and the blood flow stagnation index measured by optical coherence tomography were input to predictively calibrate the invasiveness and liquefaction boundaries of the lesion. The predicted results were used to dynamically optimize the initial tensor field to form a four-dimensional risk assessment tensor field that includes pathological trends.

[0011] Sp2, Adaptive debridement path planning and priority decision-making:

[0012] Based on the four-dimensional risk assessment tensor field, a three-dimensional debridement path is automatically planned through a multi-objective optimization algorithm that combines minimum bioturbation with maximum debridement efficiency;

[0013] Sp3, closed-loop precise debridement based on acoustic fluid coupling:

[0014] Adaptively matching the ultrasonic cutting frequency to the dentin hardness and thickness distribution in the digital twin model to achieve the highest cutting efficiency while minimizing the risk of microcracks; followed by rinsing;

[0015] Sp4, function-oriented repair and stress reconstruction:

[0016] After cleaning is completed, the system simulates the stress distribution of the tooth under different chewing forces, and combines the cumulative fatigue damage map caused to the dentin wall during the debridement process to calculate the ideal stress distribution model of the repaired tooth. The function-oriented repair and stress reconstruction include material gradient distribution repair and embedded structure repair. During the material gradient distribution repair, when it is found that the structural change of the repair section will cause the local stress to exceed 150 MPa in the dentin, the secondary stress conduction pathway is found and planned in the adjacent healthy tooth tissue. The main load is guided to the secondary pathway by adjusting the geometric shape of the restoration. The embedded structure repair uses a bionic endodontic post 3D printed with PEEK material;

[0017] Sp5, Multimodal long-term stability monitoring and self-consistent verification:

[0018] After the restoration is completed, a passive microsensor array is implanted on the surface of the crown or in the restoration for long-term monitoring of multi-source physiological parameters including bioelectrical impedance spectroscopy, temperature gradient and acoustic emission signals; a personalized health baseline associated with the patient's digital twin model is established, and the data sent back by the sensor is continuously analyzed through cloud-based algorithms.

[0019] Preferably, the time dimension of the four-dimensional risk assessment tensor field in Sp1 is constructed through a pathological evolution model based on cellular automata. The pathological evolution model based on cellular automata uses the initial blood perfusion map and lesion boundary measured by optical coherence tomography as initial conditions to simulate the predicted path and rate of pulp tissue necrosis and inflammation spread.

[0020] Preferably, the multi-objective optimization algorithm in Sp2 is an improved ant colony algorithm, in which the volatilization and update of pheromones depend on the path length, the risk assessment tensor value of the area passed by the path, the predicted debris generation amount and the proximity to the neurovascular bundle.

[0021] Preferably, the Sp3 also includes a feedback sub-component based on the acoustic fingerprint of cavitation bubbles, which identifies the states of effective cleaning and ineffective or dangerous cleaning by analyzing the acoustic signal spectrum when the cavitation bubbles generated during the ultrasonic debridement process rupture, and adjusts the ultrasonic power and frequency in real time to maximize the cleaning efficiency and inhibit excessive erosion of the root canal wall.

[0022] Preferably, in the Sp3, when the Raman spectroscopy sensor detects the presence of specific protein markers related to nerve tissue in the flushing fluid, the system immediately suspends mechanical cutting of the area and switches to a low-pressure, energy-free flushing mode to protect the nerve bundles.

[0023] Preferably, the distribution strategy of the repair in Sp4 is generated by a topology optimization algorithm, which takes maximizing the overall stiffness of the tooth after repair and minimizing stress concentration as optimization goals, and takes the physiological performance parameters of dentin as constraints.

[0024] Preferably, the bioelectrical impedance spectroscopy analysis in Sp5 is processed using a deep convolutional neural network, which is trained with data to automatically identify weak but characteristic patterns associated with microleakage, secondary caries, and root cracks from complex impedance spectra.

[0025] Preferably, the dental pulp treatment system of the oral dental pulp cleaning and treatment method comprises:

[0026] a data acquisition module, including a cone-beam computed tomography interface and an optical coherence tomography probe interface;

[0027] Computing and modeling servers, pre-installed with digital twin construction software, pathology evolution models, path planning algorithms, and prosthetic design software;

[0028] An intelligent debridement execution unit, comprising a multi-degree-of-freedom robotic arm, a PEEK end effector with an integrated ultrasonic transducer and fluid channel, and a control system and fluid pump connected to the actuator;

[0029] A real-time sensing and feedback module, including a Raman spectrometer for analyzing reflux fluid, a micro pressure sensor for monitoring apical pressure, and a hydrophone for collecting air bubble sound signals;

[0030] A prosthetic fabrication and implantation module, including a multi-material DLP 3D printer and a navigation system for guiding prosthetic implantation;

[0031] Long-term monitoring and early warning terminal, including signal reading equipment for receiving and processing passive sensor data, and a user interface for displaying analysis results and early warning information.

[0032] Preferably, a hard real-time control closed loop with a delay of less than 10 milliseconds is formed between the calculation and modeling server, the intelligent debridement execution unit and the real-time sensing and feedback module to ensure the immediacy and safety of the system response during the high-speed debridement process.

[0033] Beneficial effects

[0034] The present invention provides an oral pulp cleaning and treatment method and a pulp treatment system thereof, which have the following beneficial effects:

[0035] 1. The present invention constructs a high-fidelity digital twin model by integrating the macrostructure of CBCT with the microstructure and blood flow data of OCT, realizing for the first time visual diagnosis of the root canal at the micron level, and can accurately identify microcracks and infected blind spots. The chemical feedback closed loop based on Raman spectroscopy analysis upgrades the endpoint of debridement from the traditional "mechanical forming completion" to "biological cleaning completion". By real-time monitoring of the marker concentrations of necrotic tissue and bacterial metabolites, it provides an objective and quantitative "cleaning" indicator for the operation, ensuring the thorough removal of all infected areas. This full-process precision from diagnosis to treatment endpoint judgment overcomes the blindness of traditional technologies, can significantly improve the first-time success rate of root canal treatment, and reduce prolonged healing and secondary treatments caused by incomplete debridement.

[0036] 2. The present invention places the treatment process under strict safety monitoring through the real-time feedback closed loop of multiple sensors: the PID closed-loop control based on the apical micro-pressure sensor can always maintain the fluid pressure in the apical area below the safety threshold, effectively preventing chemical and physical stimulation of the tissues around the apex; the acoustic fingerprint analysis of cavitation bubbles collected by the hydrophone can optimize the ultrasonic power in real time, avoiding excessive cutting and damage to the dentin while ensuring debridement efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a system diagram of the present invention;

[0038] Figure 2 It is a schematic diagram of the stages of the present invention;

[0039] Figure 3 This is a diagram of a case of the present invention;

[0040] Figure 4 This is the second case diagram of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Specific embodiment one:

[0043] like Figures 1 to 2 As shown, a method for cleaning and treating oral pulp includes the following steps:

[0044] Sp1. Multimodal digital twin construction and risk prediction:

[0045] By fusing 3D hard tissue structural data acquired with a voxel resolution of 75 microns from cone-beam computed tomography (CBCT) with 10-micron depth resolution data on the soft tissue microstructure and blood flow within the dental pulp cavity (OCT), a multi-scale, high-fidelity digital twin model of the dentin-pulp complex (DPC) was constructed through a multiresolution image registration process. This process first employed a global affine transformation based on normalized mutual information for coarse registration, followed by a free-form deformation-based B-spline interpolation nonrigid registration algorithm for fine alignment. The L-BFGS optimizer was then used to minimize the registration cost. Based on a pre-defined structural tensor analysis algorithm, the gradient covariance matrix within each voxel neighborhood of the CBCT image was calculated to determine its eigenvectors and eigenvalues, thereby calibrating the 3D orientation and anisotropy of the dentinal tubules. This was then combined with a U-Net-based neural network to learn and segment the probabilistic distribution paths of neurovascular bundles from an anatomical atlas database. Ultimately, an initial tensor field representing the anisotropy of energy transfer and diffusion was generated. The multidimensional feature vector of the lesion region, composed of 12 dimensions of data, including voxel CBCT grayscale values, texture features from the gray-level co-occurrence matrix, OCT signal intensity, and Doppler shift values, is input into a pre-trained model consisting of two stacked long-short-term memory (LSTM) layers, each with 128 hidden units. This model, through a softmax output layer, probabilistically predicts the invasiveness and liquefaction boundaries of the lesion. The predictions are used to dynamically optimize the initial tensor field, forming a four-dimensional risk assessment tensor field that reflects pathological trends. The temporal dimension of the four-dimensional risk assessment tensor field in Sp1 is constructed using a cellular automaton-based pathological evolution model. The cellular space of this model is the discretized grid of the digital twin model, with cellular states including healthy, inflamed, necrotic, and fibrotic. The state transition rules are defined as follows: the probability of a healthy cell transitioning to an inflamed state, P(H→I), is proportional to the number of inflamed cells in its neighborhood and the concentration of inflammatory mediators simulated by the discrete Laplace operator. The probability of an inflamed cell transitioning to a necrotic state, P(I→N), is based on a nutrient consumption model in which nutrient levels are inversely proportional to the local inflammatory cell density and the distance to the blood supply source as determined by optical coherence tomography (OCT). Using the initial blood perfusion map and lesion boundaries measured by optical coherence tomography as initial conditions, the model simulates the predicted path and rate of spread of pulp tissue necrosis and inflammation over the next 24 to 72 hours.

[0046] Sp2, Adaptive debridement path planning and priority decision-making:

[0047] Based on a four-dimensional risk assessment tensor field, a multi-objective optimization algorithm combining minimal bioturbation with maximum debridement efficiency automatically plans a three-dimensional debridement path. The path includes not only spatial coordinates but also the recommended operation sequence for each point along the path. The path is dynamically divided into a guided intervention segment, a core lesion segment, and a dynamic reflow segment. Based on the risk level and operation sequence of each segment, the system generates a nonlinear operation priority sequence, allowing the current operation to be interrupted under specific conditions to prioritize higher-risk events. The multi-objective optimization algorithm in Sp2 is a modified ant colony algorithm. Its core is to minimize a cost function consisting of four weighted terms: Cost(P)=w1·L(P)+w2·∫Risk(s)ds+w3·∫Debris(s)ds+w4·(1 / d_min), where L(P) is the total length of path P, ∫Risk(s)ds is the integral of the four-dimensional risk tensor along the path, ∫Debris(s)ds is the total debris generation predicted based on dentin hardness and instrument parameters, and d_min is the minimum Euclidean distance between the path and the neurovascular bundle. The pheromone update rule Δτ_ij=(1-ρ)·τ_ij+ΣΔτ_ij^k, where ρ is the volatility coefficient, is implemented. The pheromone contribution Δτ_ij^k of an individual ant k is not only proportional to the inverse of the total path cost but also modulated by a local heuristic function η_ij, whose value is inversely proportional to the risk tensor of the next node j. This guides the ant colony to simultaneously pursue local risk avoidance while simultaneously pursuing global optimization.

[0048] Sp3, closed-loop precise debridement based on acoustic fluid coupling:

[0049] A debridement tool is used that is driven by ultrasonic energy and pulsed fluid in a coordinated manner. The ultrasonic cutting frequency is adaptively matched to the resonant frequency through an integrated phase-locked loop (PLL) circuit. This circuit compares the phase of the driving current with the instrument vibration displacement sensed by the piezoelectric sensor. The resulting error signal is smoothed by a low-pass filter and input to a voltage-controlled oscillator to adjust the output frequency, so that the system is always locked at the resonant point with the smallest phase difference, that is, the highest energy transfer efficiency, to achieve the highest cutting efficiency while minimizing the risk of microcracks. Afterwards, irrigation is performed, and the irrigation pressure and pulse mode are coupled with the geometry of the dynamic reflow section and the real-time fluid impedance. A proportional-integral-differential controller running at a frequency of 100 Hz is used for closed-loop control. The controller uses the measured value of the micro-pressure sensor in the apical area as input and uses u(t) to calculate the pressure. =Kp·e(t)+Ki·∫e(t)dt+Kd·de(t) / dt formula is used to calculate the output value to directly control the speed of the fluid pump, maintaining the pressure in the root apex area always below the safety threshold of 15 mmHg; the biochemical composition of the irrigation fluid is analyzed in real time by a Raman spectroscopy sensor integrated in the reflux path. The data processing process includes: using a median filter to remove cosmic rays, using asymmetric least squares method for baseline correction, and then using a database containing pure spectra of various bacterial porphyrins, hydroxyproline, human serum albumin and other substances, through classical least squares regression analysis, the concentration of each marker is calculated in real time. When the concentration of a specific marker is lower than the clearance threshold for five consecutive sampling cycles, the system determines that the area has been cleared and automatically switches to the next path point, realizing intelligent judgment based on the chemical clearance endpoint. Sp3 also includes a feedback sub-step based on the acoustic fingerprint of cavitation bubbles. By performing a short-time Fourier transform on the acoustic signal collected by the hydrophone, the acoustic signal spectrum of the cavitation bubbles generated during ultrasonic debridement is analyzed. When the signal is detected to transition from a stable harmonic peak to high-intensity broadband noise, the system identifies the transition from a stable cavitation state that effectively cleans to an ineffective or dangerous inertial cavitation state, and adjusts the ultrasonic power down by 5% to 10% in real time to maximize cleaning efficiency and inhibit excessive erosion of the root canal wall. In Sp3, when the Raman spectroscopy sensor detects the presence of specific protein markers associated with nerve tissue, such as gangliosides or myelin basic protein, in the irrigation fluid, the system immediately suspends mechanical cutting in that area and switches to a low-pressure, energy-free chemical irrigation mode with a flow rate of 0.5 ml per minute to protect the nerve bundles.

[0050] Sp4, function-oriented repair and stress reconstruction:

[0051] After cleaning is complete, the system uses finite element analysis to simulate tooth stress distribution under varying chewing forces ranging from 50 N to 500 N. Combined with the cumulative fatigue damage map of the dentin wall caused during debridement, calculated based on the Miner linear cumulative damage criterion, it calculates an ideal stress distribution model for the restored tooth. Based on this ideal model, a functional gradient restoration scheme is automatically designed, which includes both material gradient distribution restoration and embedded structural restoration. A load-bearing path optimization mechanism is introduced. If the calculation finds that structural changes in any repair segment will cause local stress to exceed the physiological limit of 150 MPa in dentin, the system automatically searches for and plans a secondary stress conduction pathway in the adjacent healthy tooth tissue and directs the primary load to this secondary pathway by adjusting the restoration's geometry. The restorative material distribution strategy in Sp4 is generated using a solid isotropic material penalty topology optimization algorithm, which iteratively solves on the finite element mesh of the restoration space. Its objective function is to minimize the total structural strain energy, subject to the constraint that the total volume of the restorative material must not exceed a preset value. In each iteration, the pseudo-density ρ∈[0,1] of each unit is updated, and the Young's modulus of the unit is defined as E(ρ)=E_min+ρ^p·(E_max-E_min), where the penalty factor p is usually 3 to drive the pseudo-density to converge to 0 or 1. The final pseudo-density distribution map is converted into a CAD model to guide the 3D printing of functional gradient materials. The embedded structure repair method designed in Sp4 is a biomimetic dental pulp post 3D printed with PEEK material, the surface of which is modified with a bioactive coating that guides the directional differentiation of dental pulp stem cells. The coating is constructed by electrostatic layer self-assembly technology, alternately depositing positively charged poly-L-lysine and negatively charged sodium alginate pre-loaded with growth factors to form 20 double-layer structures, of which the inner 10 layers are loaded with vascular endothelial growth factor and the outer 10 layers are loaded with bone morphogenetic protein-2 to achieve controllable and timed factor release, aiming to promote the regeneration and vascularization of periapical tissue.

[0052] Sp5, Multimodal long-term stability monitoring and self-consistent verification:

[0053] After the restoration is completed, a passive microsensor array is implanted on the surface of the crown or in the restoration for long-term monitoring of multi-source physiological parameters including bioelectrical impedance spectra with a frequency range of 1 Hz to 1 MHz, temperature gradients, and acoustic emission signals with a frequency of more than 100 kHz generated by microcrack expansion; a personalized health baseline associated with the patient's digital twin model is established, and the data sent back by the sensor is continuously analyzed through cloud-based algorithms. When specific signal patterns indicating structural abnormalities are detected, the system issues an early warning to doctors and patients; when the system issues a warning, it automatically retrieves the historical data of the tooth for a virtual stress test. If the five-year failure probability predicted based on the probabilistic fracture mechanics model exceeds 5%, a follow-up visit is recommended and a detailed diagnostic report is provided. Bioelectrical impedance spectroscopy analysis in Sp5 is processed using a one-dimensional deep convolutional neural network. Its architecture is as follows: an input layer receives impedance data at 256 frequencies, followed by two "convolution-activation-pooling" modules, each containing 64 and 128 convolution kernels of size 5, ReLU activation functions, and max pooling layers of size 2. A flattening layer then connects to a dense layer containing 100 neurons and a dropout layer operating at a rate of 0.5. Finally, a softmax output layer classifies the signal as "healthy," "microleakage," or "root fracture." This deep convolutional neural network was trained using a dataset of over 50,000 samples, validated and labeled using micro-CT and dye permeation methods from in vitro artificial defect models and clinical trials. This enables it to automatically identify subtle but characteristic patterns associated with microleakage, secondary caries, and root fractures from complex impedance spectra.

[0054] In summary, the above workflow is as follows: Figure 2 As shown:

[0055] Part 1 of the treatment plan:

[0056] 1. Data collection;

[0057] 1.1. The operator initiates a scan command;

[0058] 1.2. CBCT data is received, parsed, and verified via the DICOM interface;

[0059] 1.3. The OCT probe collects interference signals, and the GPU reconstructs the B-scan image sequence in real time;

[0060] 2. Construction of four-dimensional digital twins;

[0061] 2.1. Accurately register CBCT and OCT data at multiple resolutions.

[0062] 2.2. Generate a dentin structure tensor field and a neurovascular bundle probability map;

[0063] 2.3. LSTM model predicts initial lesion invasiveness and liquefaction boundary based on multimodal features;

[0064] 2.4. Cellular automata models predict future pathological trends and form a 4D risk field.

[0065] 3. Intelligent path planning;

[0066] 3.1. CUDA-accelerated ant colony algorithm for multi-objective optimization based on 4D risk field;

[0067] 3.2. Generate a debridement pathway consisting of three segments: guidance, core, and reflow, and a nonlinear priority sequence;

[0068] Part II of Intelligent Debridement and Repair:

[0069] 4. Automated debridement execution;

[0070] 4.1. The robot moves the end effector along the planned path.

[0071] 4.2. Enable the underlying hard real-time feedback loop (<5ms);

[0072] 4.3. Start the high-level chemical feedback loop (<100ms);

[0073] 5. Function-oriented repair;

[0074] 5.1. The server runs a topology optimization algorithm to design a functionally graded restoration.

[0075] 5.2. Multi-material 3D printer produces biomimetic endodontic posts based on CAD models;

[0076] 5.3. Under the real-time guidance of the optical navigation system, a robotic arm precisely implants the restoration.

[0077] Part III Long-term Monitoring:

[0078] 6. Regular data backtesting;

[0079] 6.1. The patient uses an NFC reader device to collect data from an implanted passive sensor;

[0080] 6.2. Data is securely uploaded to the monitoring terminal software via Bluetooth;

[0081] 7. Intelligent analysis and early warning;

[0082] 7.1. Terminal software automatically updates and visualizes data trend charts;

[0083] 7.2. A preloaded CNN model analyzes the bioelectrical impedance spectrum and outputs the diagnostic probability of microleakage or root fracture.

[0084] 7.3. The system integrates all data and updates the five-year failure probability. If it exceeds the preset threshold, it triggers a two-way warning for doctors and patients. Specific embodiment two:

[0086] like Figures 1 to 2 As shown in the figure, the data acquisition module begins with the operator initiating an acquisition command. It includes a DICOM C-STORE service program that continuously listens on a specific port via the TCP / IP protocol. Upon receiving a C-STORERQ request from the hospital's PACS system, the module automatically validates the DICOM header information, parses, and streams image datasets up to 2048x2048x1920 voxels and 16-bit depth to the computing server's temporary storage area. After the transfer, data integrity is verified using an MD5 checksum. The module also includes a forward-looking fiber-optic common-path optical coherence tomography probe with a diameter of 0.8 mm and a built-in MEMS scanning galvanometer. After insertion into the root canal, its laser source is activated, and the returned interferometric spectral signal (A-scan) is acquired in real time by a high-speed digitizer with a PCIe interface and a sampling rate of 200 MS / s. This signal is then transferred to the GPU memory without latency via direct memory access (DMA) technology. The dedicated CUDA kernel function running on the GPU will perform fast Fourier transform, dispersion compensation and logarithmic compression on each A-scan data stream, reconstruct it into a B-scan cross-sectional image sequence in real time, and synchronize it with the CBCT data stream to form the original input for subsequent modeling.

[0087] The computing and modeling server features a high-performance server equipped with dual-socket Intel Xeon Platinum processors, 256GB of DDR4 error-correcting code memory, and four NVIDIA A100 GPUs interconnected by NVLink. Its core workflow automatically initiates upon detecting the completion of data acquisition: First, the digital twin construction software performs anisotropic diffusion filtering on the CBCT data on the CPU to reduce noise while preserving the edge details of the dentinal tubules. Next, a pre-trained U-Net neural network performs semantic segmentation on the OCT image sequence on the GPU to accurately extract the three-dimensional surface of the dental pulp cavity. A multi-resolution registration process is then initiated, using the segmented dental pulp cavity surface as a high-confidence target. A global affine transformation coarse registration is first performed on the CPU. The result is then used as the initial condition to parallelly compute a B-spline-based non-rigid deformation field on the GPU. The L-BFGS optimizer minimizes the normalized mutual information cost function to achieve precise subvoxel-level alignment. After alignment, a GPU-accelerated structural tensor analysis module calculates the anisotropy of the entire dentin region and performs weighted fusion with the neurovascular bundle probability map segmented from the anatomical atlas by another U-Net model to generate an initial tensor field. Subsequently, the 12-dimensional feature vector extracted from the multimodal data is fed into a pre-trained model containing two stacked long short-term memory network layers loaded on the GPU to output an initial risk prediction. Finally, a cellular automaton simulation program implemented in C++ is launched on the CPU, using the LSTM prediction results and OCT blood flow map as the initial conditions at t=0. It iterates according to the reaction-diffusion equation that includes the diffusion of inflammatory mediators and nutrient consumption rules to deduce a four-dimensional risk assessment tensor field that includes future pathological trends, and stores the complete digital twin and risk data in a structured manner.

[0088] The intelligent debridement execution unit, whose workflow is initiated by priority path commands issued by the server, comprises a six-axis collaborative robotic arm with ±0.02mm repeatability. Its controller receives a sequence of path points containing six degrees of freedom (DOF) spatial positions sent by the server at 60Hz and generates smooth joint motion trajectories through inverse kinematics and cubic spline interpolation. A PEEK end effector, made of PEEK and autoclavable, integrates a PZT-8 piezoelectric ceramic ultrasonic transducer and coaxial dual fluid channels. The transducer operates in the 25-45kHz frequency range, with the inner and outer fluid channels used for irrigation and suction, respectively. The actuator is connected to a field-programmable gate array (FPGA)-based real-time control system and a precision peristaltic pump. When the robotic arm moves, the FPGA controller synchronously activates the peristaltic pump and ultrasonic generator and immediately initiates a built-in phase-locked loop (PLL) circuit. This circuit automatically finds the optimal resonant frequency for the current dentin structure within milliseconds by comparing the phase difference between the drive current and the vibration signal fed back by the piezoelectric sensor, maximizing energy transfer efficiency. During the entire debridement process, the FPGA controller independently and quickly executes underlying safety and efficiency assurance tasks.

[0089] The real-time sensing and feedback module, which forms the core of the hard real-time closed loop and operates with uninterrupted data flow, includes a compact Raman spectrometer integrated with the end effector via optical fiber and using a 785-nanometer excitation light source. Its CCD sensor acquires spectra at a frequency of 20 Hz. The collected raw data is processed by a dedicated digital signal processor for cosmic ray removal, baseline correction, and classical least squares solution. The calculated chemical composition concentration vector is then sent to the computing server via Ethernet for high-level decision-making. A fiber-optic micro-pressure sensor based on the Fabry-Perot interferometry principle, integrated at the top of the PEEK end effector, outputs the decoded pressure value as an analog voltage signal through its optical interrogator. Furthermore, a micro-needle hydrophone with a bandwidth of up to 20 MHz, integrated in the return suction line, outputs analog acoustic signals covering the full spectrum of cavitation activity. The analog outputs of pressure and acoustic signals are directly fed into the FPGA real-time control system of the intelligent debridement execution unit and digitized by its internal high-speed analog-to-digital converter. The digitized data are directly used as input for the PID pressure control algorithm and the FFT acoustic fingerprint analysis algorithm. The calculation results of these two algorithms directly generate control instructions for the peristaltic pump motor and ultrasonic generator power, forming a complete underlying hardware-level feedback loop with a delay of less than 5 milliseconds.

[0090] The restoration fabrication and implantation module initiates its manufacturing process after the restoration design is completed on the server. This process involves a multi-material DLP 3D printer with a print resolution of 15 microns, utilizing multi-material digital light processing technology. The server sends the STL model of the functionally graded restoration to the printer, where built-in software slices the model and assigns precise material mix ratios to each system. During printing, the DLP projector dynamically adjusts the illumination pattern based on the material distribution map of each layer, selectively curing the different resin mixtures. Layer by layer, the restoration is formed and then enters an automated cleaning and UV post-curing unit. The implantation process is as follows: the patient wears a dynamic reference dental arch splint equipped with a tracer ball. An optical spatial navigation system, comprised of a binocular infrared camera, first tracks the target point by tapping the probe on the tooth surface, precisely aligning the patient's dental arch with the preoperative CBCT data. The printed restoration is then mounted on a dedicated fixture at the end of a robotic arm, where the tracer ball is recognized and registered. At this time, the system screen can display the virtual position of the restoration relative to the three-dimensional model of the tooth in real time. Under the guidance of the navigation system, the operator controls the robotic arm along the preset implantation path and accurately places the restoration into the cleaned root canal with sub-millimeter accuracy.

[0091] The long-term monitoring and early warning terminal, designed for periodic or on-demand inspections, consists of a handheld NFC signal reader that wirelessly communicates with an implanted passive sensor via the Near Field Communication (NFC) protocol. During an inspection, the reader's antenna coil generates an electromagnetic field to power the sensor. Once awakened, the sensor then performs a bioelectrical impedance spectrum scan, temperature measurement, and acoustic emission event monitoring. The collected encrypted data packets are transmitted back to the reader using load modulation technology. The terminal also features a graphical user interface running on a tablet or workstation, with the reader syncing data to the software via Bluetooth. The software first decrypts and verifies the data, then immediately updates the historical trend chart with new data points. The latest impedance spectrum data is then fed into a pre-loaded deep convolutional neural network model consisting of three one-dimensional convolutional layers and two fully connected layers for analysis. The model's classification results and a recalculated five-year probability of failure based on the new data are displayed instantly on the terminal's main interface dashboard, completing a complete health status assessment and early warning process.

[0092] The computation and modeling server, intelligent debridement execution unit, and real-time sensing and feedback module form a hard real-time control loop with data transmission and processing latency of less than 10 milliseconds. The specific path of this closed loop is as follows: Signals collected by the pressure sensor and hydrophone are processed by the FPGA real-time control system through PID and phase-locked loop operations, directly generating control instructions for the peristaltic pump and ultrasonic generator. The response time of this low-level safety loop is guaranteed to be within 5 milliseconds. Simultaneously, real-time sensor data is uploaded to the computation and modeling server in the form of data packets. The server makes macro-decisions based on Raman spectroscopy analysis and path planning, and sends the updated path instructions to the robotic arm controller. The entire high-level feedback loop period is strictly limited to 10 milliseconds to ensure the immediacy and absolute safety of the system response during high-speed debridement. Specific embodiment three:

[0094] Based on the technical solutions of Specific Example 1 and Specific Example 2, further case supplementary explanation is given:

[0095] like Figure 3 Case 1 shown: Complex debridement and restoration of a C-shaped root canal of a mandibular second molar. The patient presented with a history of occlusal discomfort and mild discomfort with hot food in the lower right posterior tooth. Figure 3 From left to right, (1), (2), and (3) are the visualized images after processing by CBCT, OCT, and the technical solution of the present application, respectively. The left side shows the original images from multiple sources, each of which has its inherent advantages and disadvantages, which together constitute a complex diagnosis;

[0096] Conventional X-rays revealed a fuzzy, low-density shadow around the apex of the mandibular right second molar (#47). A pulp vitality test was negative, leading to a preliminary diagnosis of chronic apical periodontitis. The clinical challenge lies in the high incidence of C-shaped root canals in mandibular second molars. Their complex anatomy, including isthmuses, fin-shaped communicating branches, and irregular depressions, makes thorough debridement difficult with traditional instruments, a high-risk factor for conventional root canal treatment failure. The long-term monitoring and early warning terminal also includes graphical user interface software and an implantable passive multimodal sensor. The graphical user interface software incorporates a 3D data visualization module, a 1D CNN diagnostic analysis engine, and an early warning and report generator.

[0097] Sp1. Accurate diagnosis and four-dimensional modeling: The system first integrated CBCT and OCT data. CBCT clearly outlined the typical "C"-shaped fusion of the root of tooth #47, but the internal details were blurred. Subsequently, the operator inserted the 0.8mm OCT probe into the pulp cavity, and the system scanned and reconstructed a high-resolution three-dimensional image of the root canal system in real time. The OCT image clearly revealed the presence of a large amount of signals that appeared to be necrotic tissue in the wide isthmus connecting the buccal and lingual main root canals, and marked two fin-shaped communicating branches that were extremely difficult to reach with traditional instruments. Based on these fine structures, the system's four-dimensional risk assessment tensor field marked these areas as "high-risk biofilm-enriched areas" and predicted that their liquefaction boundaries would tend to expand toward the root apex within the next 48 hours.

[0098] Sp2. Multi-Objective Path Planning: Planning is initiated using an ant colony algorithm on the computing server. The algorithm's objective function is not only to minimize the path but, more importantly, to maximize coverage of the C-shaped isthmus and fin-shaped communicating branches, while strictly avoiding the weakest areas of the root canal wall. Ultimately, the system generates not a single straight path but a "sweeping" three-dimensional debridement path that covers the entire C-shaped system, with different operational priorities and recommended energy parameters set for different areas (main canal, isthmus, communicating branches).

[0099] Sp3, Closed-loop Precision Debridement: The intelligent debridement execution unit starts working. The robotic arm guides the end effector to move along the "sweep path". When entering the wide isthmus, the phase-locked loop (PLL) circuit automatically adjusts the ultrasonic frequency to the resonant frequency that can stimulate the maximum cavitation effect, so as to perform efficient non-contact acoustic flow debridement. At the same time, the Raman spectroscopy sensor in the reflux fluid continuously monitors the porphyrin signal intensity representing bacterial metabolites. The system found that even after the instrument left, the porphyrin signal in the isthmus area still rebounded slightly, indicating the presence of deep biofilm. The system then instructs the robotic arm to perform a second, low-power acoustic flow cleaning of the area until the porphyrin signal is lower than the clearance threshold during the continuous sampling period, truly achieving "chemical" cleaning.

[0100] Sp4. Function-Oriented Restoration: After cleaning, the system enters the restoration design phase. Considering the thin and irregular walls of the C-shaped root canal, a topology optimization algorithm was activated to maximize fracture resistance, resulting in a non-circular biomimetic endodontic post designed to perfectly match the complex root canal morphology. The post is solidified in areas requiring support and gridded in non-load-bearing areas to conserve material and enhance interlocking with the sealant. A multi-material 3D printer was used to print the restoration using a highly resilient nanocomposite resin.

[0101] Results: The patient recovered well after surgery, and the discomfort caused by hot food disappeared. Postoperative CBCT showed that the root canal system was tightly filled with no dead corners.

[0102] Sp5. Stability Feedback: Passive sensors were implanted in the restoration. During follow-up examinations at 6 and 12 months after surgery, Mr. Zhang used an NFC handheld device to collect data. The terminal software showed that his bioelectrical impedance spectrum was highly consistent with his immediate postoperative baseline. The CNN diagnostic engine did not detect any characteristic patterns associated with microleakage, objectively confirming the long-term sealing and stability of the restoration.

[0103] OCT-CBCT fusion modeling reveals complex anatomical details that cannot be displayed by traditional imaging, while the chemical feedback closed loop based on Raman spectroscopy ensures thorough debridement of these complex areas, fundamentally solving the problem of C-shaped root canal treatment.

[0104] like Figure 4 As shown in the figure, Case 2: Diagnosis and internal retention restoration of incomplete root fracture of maxillary anterior tooth after trauma. The main complaint was that the upper left central incisor (#21) had been traumatized one year ago. After root canal treatment at an external hospital, the patient gradually developed occlusal pain and recurrent gingival abscesses in the past six months. Figure 4 From left to right, the images are CBCT, OCT, and processed using this technology. The CBCT image on the left shows the "J"-shaped bone defect of the root fracture. The OCT image in the center penetrates deep into the root canal, clearly capturing the longitudinal microcracks in the dentin wall at high resolution. The processed image on the right, after fusing the CBCT and OCT data, uses the built-in AI segmentation algorithm to automatically and precisely depict realistic and clear anatomical structures on the low-quality CBCT image, automatically overlaying a translucent color model.

[0105] CBCT scan revealed a J-shaped bone defect at the apex of root #21, a typical indirect sign of a vertical root fracture (VRF), but the fracture line itself was not visible. Conventional management involves extraction of the affected tooth. The patient was young and desired to preserve the natural tooth as much as possible. The clinical challenge was to definitively diagnose VRF intraoperatively and explore the possibility of tooth-sparing treatment.

[0106] Sp1. Qualitative microscopic diagnosis: After removing the existing root canal filling, the OCT probe was introduced into the root canal for a 360-degree scan. In the middle of the buccal wall of the root canal, the OCT image clearly revealed a microcrack line approximately 0.4 mm deep and 20 microns wide. This crack did not extend through the entire length of the root. This direct evidence confirmed the diagnosis of an incomplete vertical root fracture. The system's risk model labeled the crack line area as a "very high mechanical risk zone."

[0107] Sp2 & Sp3, Ultra-Low Disturbance Debridement: The system plans a "zero-contact" debridement path, ensuring the instrument tip always maintains a safe distance from the tube wall opposite the crack. During the debridement process, the acoustic fingerprint of cavitation bubbles collected by the hydrophone is used for real-time monitoring. If inertial cavitation characteristics (high-intensity broadband noise) that may exacerbate the crack are detected, the FPGA real-time control system instantly reduces the ultrasonic power (in less than 5 milliseconds), ensuring effective disinfection without causing secondary damage to the affected tooth.

[0108] Sp4, internal retention restoration of cracks: This is the core of the tooth-saving attempt. The topology optimization algorithm designed a special restoration solution with the primary goal of "limiting stress concentration at the crack tip." This solution is not a traditional post and core, but an "internal retention splint": on the inside of the restoration corresponding to the crack line, the algorithm specifies the use of a flexible composite material with high fracture toughness and low elastic modulus; while in other parts of the tooth, a high-strength rigid composite material is used. The 3D printer creates this functionally gradient restoration, which, after implantation, can absorb and disperse the bite force like a built-in "shock absorber," preventing the crack from continuing to expand.

[0109] Sp5. Acoustic Emission "Sentinel" Monitoring: Given the high risk of VRF, a monitoring array containing acoustic emission sensors was implanted in the crown restoration of #21. During each follow-up, the system requires Xiao Li to perform a bite action of a specified intensity. If the crack tip expands even at the micron level, the resulting stress wave will be captured by the AE sensor and recorded in the form of a high-frequency acoustic emission event. Monitoring one year after the operation showed that the AE event count was always zero, proving that the "internal retention splint" restoration scheme effectively controlled the expansion of the crack.

[0110] OCT's microscopic diagnostic capabilities transform VRF from a "guess" to a "confirmed" diagnosis, providing a basis for decision-making for optimal tooth preservation. The topologically optimized functionally graded restoration and long-term monitoring using acoustic emission sensors together form a complete "diagnosis-restoration-monitoring" closed loop, minimizing the risk of failure.

[0111] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further restrictions, an element defined by the statement "comprising a reference structure" does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0112] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for cleaning and treating oral pulp, characterized by: The following steps are involved: Sp1. Multimodal digital twin construction and risk prediction: By fusing three-dimensional structural data of dental hard tissue obtained by cone-beam computed tomography with the microstructure and blood perfusion data of the soft tissue inside the dental pulp cavity obtained by optical coherence tomography, a multi-scale, high-fidelity digital twin model of the dentin-pulp complex was constructed. The imaging features of the lesion area and the blood flow stagnation index measured by optical coherence tomography were input to predictively calibrate the invasiveness and liquefaction boundaries of the lesion. The predicted results were used to dynamically optimize the initial tensor field to form a four-dimensional risk assessment tensor field that includes pathological trends. Sp2, Adaptive debridement path planning and priority decision-making: Based on the four-dimensional risk assessment tensor field, a three-dimensional debridement path is automatically planned through a multi-objective optimization algorithm that combines minimum bioturbation with maximum debridement efficiency; Sp3, closed-loop precise debridement based on acoustic fluid coupling: Adaptively matching the ultrasonic cutting frequency to the dentin hardness and thickness distribution in the digital twin model to achieve the highest cutting efficiency while minimizing the risk of microcracks; followed by rinsing; Sp4, function-oriented repair and stress reconstruction: After cleaning is completed, the system simulates the stress distribution of the tooth under different chewing forces, and combines the cumulative fatigue damage map caused to the dentin wall during the debridement process to calculate the ideal stress distribution model of the repaired tooth. The function-oriented repair and stress reconstruction include material gradient distribution repair and embedded structure repair. During the material gradient distribution repair, when it is found that the structural change of the repair section will cause the local stress to exceed 150 MPa in the dentin, the secondary stress conduction pathway is found and planned in the adjacent healthy tooth tissue. The main load is guided to the secondary pathway by adjusting the geometric shape of the restoration. The embedded structure repair uses a bionic endodontic post 3D printed with PEEK material; Sp5, Multimodal long-term stability monitoring and self-consistent verification: After the restoration is completed, a passive microsensor array is implanted on the surface of the crown or in the restoration for long-term monitoring of multi-source physiological parameters including bioelectrical impedance spectroscopy, temperature gradient and acoustic emission signals; a personalized health baseline associated with the patient's digital twin model is established, and the data sent back by the sensor is continuously analyzed through cloud-based algorithms.

2. The method for cleaning and treating oral pulp according to claim 1, characterized in that: The time dimension of the four-dimensional risk assessment tensor field in Sp1 is constructed through a cellular automaton-based pathological evolution model. The cellular automaton-based pathological evolution model uses the initial blood perfusion map and lesion boundary measured by optical coherence tomography as initial conditions to simulate the predicted path and rate of pulp tissue necrosis and inflammation spread.

3. The method for cleaning and treating oral pulp according to claim 1, characterized in that: The multi-objective optimization algorithm in Sp2 is an improved ant colony algorithm, in which the volatilization and update of pheromones depend on the path length, the risk assessment tensor value of the area passed by the path, the predicted amount of debris generated, and the proximity to the neurovascular bundle.

4. The method for cleaning and treating oral pulp according to claim 1, characterized in that: The Sp3 also includes a feedback sub-component based on the acoustic fingerprint of cavitation bubbles. By analyzing the acoustic signal spectrum when the cavitation bubbles generated during ultrasonic debridement burst, it identifies the states of effective cleaning and ineffective or dangerous cleaning, and adjusts the ultrasonic power and frequency in real time to maximize the cleaning efficiency and inhibit excessive erosion of the root canal wall.

5. The method for cleaning and treating oral pulp according to claim 1, characterized in that: In the Sp3, when the Raman spectroscopy sensor detects the presence of specific protein markers related to nerve tissue in the flushing fluid, the system immediately suspends mechanical cutting of the area and switches to a low-pressure, energy-free flushing mode to protect the nerve bundles.

6. The method for cleaning and treating oral pulp according to claim 1, characterized in that: The distribution strategy of the restoration in Sp4 is generated by a topology optimization algorithm, which takes maximizing the overall stiffness of the restored tooth and minimizing stress concentration as optimization goals, and takes the physiological performance parameters of dentin as constraints.

7. The method for cleaning and treating oral pulp according to claim 1, characterized in that: The bioelectrical impedance spectroscopy analysis in Sp5 is processed using a deep convolutional neural network, which is trained with data to automatically identify weak but characteristic patterns associated with microleakage, secondary caries, and root cracks from complex impedance spectra.

8. A dental pulp treatment system for implementing the oral dental pulp cleaning and treatment method according to any one of claims 1 to 7, characterized in that: include: a data acquisition module, including a cone-beam computed tomography interface and an optical coherence tomography probe interface; Computing and modeling servers, pre-installed with digital twin construction software, pathology evolution models, path planning algorithms, and prosthetic design software; An intelligent debridement execution unit, comprising a multi-degree-of-freedom robotic arm, a PEEK end effector with an integrated ultrasonic transducer and fluid channel, and a control system and fluid pump connected to the actuator; A real-time sensing and feedback module, including a Raman spectrometer for analyzing reflux fluid, a micro pressure sensor for monitoring apical pressure, and a hydrophone for collecting air bubble sound signals; A prosthetic fabrication and implantation module, including a multi-material DLP 3D printer and a navigation system for guiding prosthetic implantation; Long-term monitoring and early warning terminal, including signal reading equipment for receiving and processing passive sensor data, and a user interface for displaying analysis results and early warning information.

9. The dental pulp treatment system according to claim 8, characterized in that: The computing and modeling server, the intelligent debridement execution unit and the real-time sensing and feedback module form a hard real-time control closed loop with a delay of less than 10 milliseconds to ensure the immediacy and safety of the system response during the high-speed debridement process.

Images