Intelligent system for laser navigation of primary tooth root canal preparation

Abstract

The application discloses a kind of intelligent preparation systems of laser navigation milk teeth root canal, it is related to milk teeth root canal treatment field, including: image processing module, self-checking registration module, structure acquisition module, spectral processing module, anomaly detection module, fusion determination module, compensation tracking module, digital modeling module, simulation learning module, trajectory generation module, execution monitoring module, safety protection module, interactive command module and record audit module;The application can more accurately distinguish infected dentin and healthy dentin, reduce the risk of misjudgment or missing infected areas, effectively reduce the risk of root tip penetration and root tip tissue damage, so that the laser cutting process always revolves around the real infected area, both can fully remove infected tissue, and can maximize the retention of healthy dental structure, improve root canal preparation efficiency and safety, reduce the dependence of doctors on experience judgment, so that milk teeth root canal preparation operation is more standardized and repeatable.

Description

Technical Field

[0001] This invention relates to the field of primary tooth root canal treatment, and more specifically to a laser-guided intelligent primary tooth root canal preparation system. Background Technology

[0002] The core step in primary tooth root canal treatment is root canal preparation, the quality of which directly affects the subsequent disinfection effect and the sealing of the root canal filling. However, due to the high individual variability and complex spatial structure of the primary tooth root canal system, many challenges remain in clinical practice. For example, infected dentin within the root canal wall is often irregularly distributed, making it difficult for traditional mechanical instruments to accurately identify its extent; at the same time, over-preparation may lead to excessively thin root canal walls or even perforation of the root apex, resulting in periapical tissue damage. In recent years, laser technology has been gradually introduced into the field of oral treatment. Lasers can achieve precise removal of tooth tissue through high energy density and have good bactericidal effects. However, existing laser treatment equipment usually lacks real-time navigation and tissue recognition capabilities, making it difficult for dentists to dynamically adjust laser energy and cutting paths according to tissue properties, which is particularly critical in primary tooth root canal preparation. Therefore, there is an urgent need to develop a laser-guided intelligent primary tooth root canal preparation system.

[0003] Existing laser-guided intelligent root canal preparation systems for primary teeth cannot accurately distinguish between infected dentin and healthy dentin, resulting in frequent misjudgments or omissions of infected areas. Furthermore, they carry a high risk of apical perforation and periapical tissue damage. Additionally, they cannot guarantee that the laser cutting process always revolves around the actual infected area, thus reducing the efficiency and safety of root canal preparation. This leads to excessive reliance on the dentist's experience and hinders the standardization and repeatability of primary tooth root canal preparation procedures. Therefore, we propose a laser-guided intelligent root canal preparation system for primary teeth. Summary of the Invention

[0004] The purpose of this invention is to overcome the deficiencies in the existing technology and provide a laser-guided intelligent root canal preparation system for deciduous teeth.

[0005] This invention proposes a laser-guided intelligent root canal preparation system for deciduous teeth, the system comprising: an image processing module, a self-registration module, a structure acquisition module, a spectral processing module, an anomaly detection module, a fusion judgment module, a compensation tracking module, a digital modeling module, a simulation learning module, a trajectory generation module, an execution monitoring module, a safety protection module, an interactive command module, and a record auditing module; The image processing module receives and preprocesses preoperative image data, initially segments the periapical structure, and generates a reference coordinate system. The self-test registration module is used to perform power-on self-tests, synchronization pulse timing, and delay corrections on various hardware devices, and to establish a unified timestamp and trigger schedule. The structure acquisition module receives the registered hardware device signal and reference coordinate system, and performs high-frame-rate three-dimensional scanning of the root canal wall, outputting high-resolution microscopic morphological data. The spectral processing module is used to capture plasma emission spectra and extract elemental characteristic information from the raw emission spectra based on a triggering schedule. The anomaly detection module collects periapical photoacoustic signals, performs time-frequency analysis and spectral feature extraction, and identifies early warning spectral patterns of penetration or tissue stimulation in real time. The fusion judgment module receives and fuses microscopic morphological data, elemental feature information, and early warning spectrum information, and outputs the three-dimensional cutting envelope of the child's deciduous teeth in real time. The compensation tracking module receives the handheld device's pose, reference coordinate system, and image information in real time, and compensates for the child's jaw movements and corrects the handheld device's pose based on the information. The digital modeling module constructs a simulation model of the child's deciduous teeth based on preoperative image data, real-time image data, three-dimensional cutting envelope, and handheld device pose. The simulation learning module runs multiple candidate execution strategies on the child's deciduous teeth simulation model and generates a strategy execution list; The trajectory generation module calculates the optical phase mask and outputs the successively triggered optical settings and scanning trajectory based on the strategy execution list and the three-dimensional cutting envelope. The execution monitoring module drives the femtosecond laser to perform cutting based on the generated scanning trajectory, and collects various data in real time for online analysis; The security protection module receives real-time spectral data, image data, simulation data, and analysis results, performs online risk assessment, selects to trigger multimodal early warning, records events, and rolls back to security parameters. The interactive command module is used to display various data during the operation, while the operator can manually select or overwrite suggestions and record signature confirmation. The record audit module is used to summarize all data from this surgery and write them into the experience database, while generating an auditable postoperative report and review data.

[0006] As a further aspect of the present invention, the specific steps of the image processing module in initially segmenting the periapical structure and generating a reference coordinate system are as follows: S1.1: Read preoperative image data from imaging equipment or PACS, then parse the header information of each image data and extract the pixel spacing, slice thickness, number of pixel rows and columns and direction cosine matrix of each image data. Stack each image data into a three-dimensional array according to its pixel position and slice spacing, and record the original scanner coordinate system information. Then, determine the uniform target voxel size according to clinical needs, calculate the corresponding position of each point on the target voxel grid in the original scanner coordinate system, and generate the volume data after resampling of each image data through trilinear interpolation, and record the new pixel spacing and volume dimension. S1.2: Extract local sub-blocks of a preset size centered on each pixel position after each resampling. Simultaneously calculate the similarity between each local sub-block and each sub-block within the preset search window. Assign corresponding weights based on the sub-block similarity. Then, replace the center pixel value by performing a weighted average on each local sub-block to remove noise from each image data. Next, use low-pass filtering to calculate the logarithmic bias field of each pixel and perform normalization correction on each pixel based on the obtained logarithmic bias field. S1.3: Calculate the grayscale histogram of the corresponding image data using each corrected pixel. Based on the grayscale distribution of the grayscale histogram, divide the image data into candidate classes of background, dentin, and pulp using the Otsu multi-threshold method. Then, perform three-dimensional connected component analysis on each candidate class and select connected components whose volume and shape indicators conform to the root canal characteristics. S1.4: After centering the pixel set of each connected component, construct its covariance matrix. Then, determine the initial principal axis direction of the root canal through principal component analysis and output the initial principal axis direction of the root canal in the form of point + direction vector. At the same time, calculate the signal-to-noise ratio and contrast index of the ROI region in the image data, and evaluate the current image data quality based on the calculation results. If the image data quality does not meet the preset standard, prompt whether it is necessary to re-import the image data or adjust the pixel size. Record and output the pixel spacing, volume dimension, coordinate system transformation matrix and initial root canal axis.

[0007] As a further aspect of the present invention, the specific calculation formula for the trilinear interpolation described in S1.1 is as follows:

[0008] In the formula, This represents the target grid points after resampling. Volume data; This represents the intensity values ​​of the eight neighboring corners surrounded by the resampled points within the original volume; Represents the relative position weight along the x-axis; Represents the relative position weight along the y-axis; Represents the relative position weight along the z-axis.

[0009] As a further aspect of the present invention, the specific steps of the structure acquisition module to perform high-frame-rate three-dimensional scanning of the root canal wall and output high-resolution microscopic morphological data are as follows: S2.1: Install the sample arm and reference arm according to the predetermined probe geometry, record the initial estimate of the physical optical path difference between the sample and the reference arm, and simultaneously acquire the original spectrum of the calibration white plate or reference mirror. Then, smooth and correct the instrument response of the original spectrum to obtain the corrected reference spectrum. Based on the corrected reference spectrum, construct a set of index mapping tables, and resample the original sampling points to k points at equal intervals. Then, interpolate each wavelength based on the same index mapping table to obtain the resampled spectral sequence. S2.2; Remove the DC component from each resampled spectral sequence, and calculate the wavenumber-dependent compensation phase curve by phase fitting for each wavelength and save it as a compensation phase term. Then, perform phase preprocessing on the processed spectral sequence using the calculated compensation phase term, and then perform discrete inverse Fourier transform on the preprocessed spectral sequence to obtain the complex-valued depth spectrum. After that, perform envelope detection on the complex-valued depth spectrum to obtain the amplitude-type A-scan curve and record it as a depth profile. S2.3: Calculate the time required for single volume acquisition and match it with the acceptable range of motion in the child's oral cavity. After matching, use the lateral scanning mechanism to perform multiple scans in the lateral direction at preset intervals to form a set of two-dimensional lateral-depth planes. Repeat the acquisition of the lateral-depth planes and acquire three-dimensional volumes at different lateral angles. At the same time, write the A-scan curve into the volume buffer in frame form. S2.4: Based on the cross-correlation method, the adjacent transverse-depth planes in the child's oral cavity are compared to obtain the corresponding displacement estimates. Then, the displacement estimates calculated for each frame are inversely transformed in the volume buffer to compensate for the structural displacement caused by the motion. S2.5: The 3D volume obtained by discrete inverse Fourier transform is subjected to local contrast enhancement, multi-angle or multi-view composite despeccization, and small-scale sharpening filtering to enhance the details of the tube opening and microcracks. After processing, the depth section, surface point cloud, and local magnified view are exported.

[0010] As a further aspect of the present invention, the specific steps of the anomaly detection module in real-time identifying the warning spectrum of penetration or tissue stimulation are as follows: S3.1: Select the pulse energy and pulse width, and measure the effective irradiation area of ​​the incident light in the local area. Then, use a low-energy short pulse or narrow-band continuous wave probe light to irradiate the periapical region at the predetermined detection position using an ultrasonic microsensor. Calculate the local light flux and the initial light energy density absorbed by the tissue. Record all parameters for each detection. Based on the known physical relationship and the absorbed energy density, calculate the initial ultrasonic pressure amplitude caused by local instantaneous absorption and record its spatial distribution estimate. S3.2: Calculate the expected arrival time based on the geometric position of the sensor array and the distance between the generation point and the ultrasonic microsensor, and record the expected arrival time of each ultrasonic microsensor as an initial alignment reference. Then, record the original time-domain waveforms acquired by each ultrasonic microsensor and attach the corresponding timestamps. Finally, perform pre-amplification, sampling rate setting and ADC quantization processing on each original time-domain waveform. S3.3: Perform anti-aliasing filtering on each processed time-domain waveform, then select the corresponding bandpass frequency band based on the sensor bandwidth and target spectrum range, and perform bandpass filtering on it. After that, based on the obtained initial alignment reference, perform normalized gain correction and time alignment on the time-domain waveform of each sensor transmission channel. S3.4: By using short-time Fourier transform, the energy distribution of each filtered and aligned time-domain waveform in the time-frequency plane is obtained. Based on a preset sliding time window, multiple sets of spectral features of each time-domain waveform in different time periods are extracted. Then, each set of spectral features is compared with the pre-calibrated corresponding hazard spectrum reference statistics to obtain the corresponding risk score. Then, the Mahalanobis distance function is used to compress the multidimensional features into a single risk value. If the risk value exceeds the set threshold, it is marked as close to penetration or tissue stimulation within milliseconds, and the time window and corresponding features of this trigger are output.

[0011] As a further aspect of the present invention, the descriptive spectral features described in S3.4 specifically include spectral centroid, spectral energy ratio, instantaneous bandwidth, spectral peak position, and spectral energy abrupt change rate, etc.

[0012] As a further aspect of the present invention, the specific steps of the fusion determination module in real-time outputting the three-dimensional cutting envelope surface of the child's deciduous teeth are as follows: S4.1: Project the microscopic morphological data, elemental characteristic information, and early warning spectrum information of each modality onto a unified three-dimensional grid. Use multiple sets of aligned feature points to perform coarse registration on each modality data. Then, use the local least squares method to adjust each modality data to a consistent pixel level to generate the three-modal raw observations at each pixel position on the unified reference grid. After that, extract the morphological features, chemical element ratio vectors, and spectral features of each pixel on the three-dimensional grid. And obtain the infection probability of each modality through Logistic mapping. S4.2: Calculate the modal confidence value for each pixel, normalize each modal confidence value to form a corresponding weighting coefficient, and based on the obtained weighting coefficient, fuse the infection probabilities of each modality into a unified infection probability score. Then, generate the corresponding fused probability field based on the infection probability score of each pixel. S4.3: The fusion probability field is converted into a scalar field, and an initial isosurface is extracted in the pixel space according to a preset threshold. This is then used as the infection boundary. The initial isosurface is then topologically repaired, and the surface of the repaired isosurface is smoothed by Laplacian smoothing to obtain the three-dimensional cutting envelope of the child's deciduous teeth. At the same time, the geometry of the corresponding boundary surface and its corresponding local confidence are output.

[0013] As a further aspect of the present invention, the specific steps for the digital modeling module to construct a simulation model of the child's deciduous teeth are as follows: S5.1: Collect the pixel data set of the preoperative image data and the pixel data set of the intraoperative real-time image data, and select the corresponding feature points from the two sets of pixel data to perform the initial estimation of rigid registration. Then, perform fine registration on the initial estimation by iterative minimization, and transform the preoperative image data, real-time image data and three-dimensional cutting envelope into the same reference system through rigid registration. S5.2: The rigidly registered boundary surface is used as the outer surface and represented in the form of a triangular mesh. At the same time, the topology of the outer surface is repaired. A volume mesh is generated inside the outer surface using a Delaunay-based volume meshing method. Meanwhile, a locally refined mesh is generated for the root canal passage. Then, the topology, geometry and hierarchy of each mesh element are recorded, and a finite element mesh is generated. S5.3: Interpolate elemental feature information to grid nodes or cell centers, set physical-empirical mapping rules based on chemical composition and structural indicators, and convert chemical composition observations into cell-level material parameters based on the mapping rules. Simultaneously, write the cell-level material parameters into the finite element mesh, select the corresponding physical model according to clinical needs, and then perform finite element discretization on the finite element mesh to generate a simulation model of the child's deciduous teeth. Simulation prediction is then performed on the child's deciduous teeth simulation model at a preset time step. S5.4: The child's deciduous teeth simulation model receives the handheld device pose and the latest image data at each sampling time. At the same time, it maps the received data into the coordinate system of the child's deciduous teeth simulation model through registration transformation. Then, it uses a recursive state estimator to fuse the latest data with the corresponding simulation prediction results and corrects the current field state of the child's deciduous teeth simulation model through incremental correction.

[0014] As a further aspect of the present invention, the specific calculation formula for fine registration in S5.1 is as follows:

[0015] In the formula, The first one representing the preoperative imaging data Selected feature points; Represents the corresponding feature points in the real-time image data under the same matching index; ROT represents the 3×3 rigid rotation matrix to be determined; DISP represents the 3×1 translation vector to be determined; N represents the number of feature point pairs used for registration; The registration cost function is represented by an iterative minimization method to obtain the optimal ROT and DISP.

[0016] As a further aspect of the present invention, the specific steps of the simulation learning module running multiple candidate execution strategies on a child's deciduous tooth simulation model and generating a strategy execution list are as follows: S6.1: Each set of preset candidate execution strategies is represented by a corresponding parameterized variable, and each parameter is packaged into a set of column vectors as a description of the corresponding candidate execution strategy. Then, a set of basis functions is selected to describe the cross-sectional shape of the optical cutting envelope, and the three-dimensional envelope approximation of the optical field of any candidate light sheet is generated by linear combination of basis functions. S6.2: Based on available computing resources, each candidate execution strategy is mapped to a parallel simulation task in batches, and the light energy injection function of each parallel task is instantiated in the simulation model of the child's deciduous teeth. Simulation time windows and step sizes are assigned to each task, and the parameters and simulation identifiers of each task are recorded. Then, based on the three-dimensional envelope approximation, the set pulse energy and repetition pattern, the spatiotemporal distribution of energy injection density is constructed for each parallel task. S6.3: If the execution strategy is a pulse sequence, the instantaneous energy distribution corresponding to each pulse is superimposed in the time domain. If it is a continuous progression, it is mapped to a smooth time function. At the same time, the instantaneous injection in the simulation process is averaged in a short time to generate the corresponding heat source term. Then, in each parallel simulation, the corresponding heat source term is used to drive the heat conduction equation to solve the instantaneous temperature field. S6.4: Based on the instantaneous temperature field and energy absorption, a corresponding cutting depth field is established. Then, the equivalent nodal force generated by plasma expansion or impact is used as an external load, and the displacement response of the root tip and adjacent tissues is calculated to obtain the root tip disturbance measurement. After each simulation task is completed, various indicators are extracted from the solution field. S6.5: The acquired indicators are normalized and mapped to corresponding performance vectors. Each performance vector is scored according to the multi-objective aggregation criterion. Based on the score and a preset quantity threshold, a candidate strategy list and the measurement details of each strategy are output. Then, the parameters of each execution strategy in the candidate strategy list are randomly perturbed, and the model simulation is repeated to generate the corresponding score statistical distribution. The expected score and score standard deviation of each candidate strategy are calculated. Based on the preset robustness index, the candidates are reordered, and the set of priority strategies ordered by robustness is output.

[0017] As a further aspect of the present invention, the specific calculation formula for the linear combination of basis functions described in S6.1 is as follows:

[0018] In the formula, Represented by parameters The generated cutting envelope indicator function; Representative and simulation identifiers; Represents strategy parameters; Representing the A pre-selected basis function.

[0019] As a further aspect of the present invention, the specific calculation formula for the linear combination of basis functions described in S6.1 is as follows: In the formula, represents the cutting envelope indicator function generated by the parameters; represents the simulation identifier; represents the strategy parameters; and represents the th pre-selected basis function.

[0020] As a further aspect of the present invention, the parameterized variables in S6.1 specifically include the geometric shape coefficient of the light sheet or light ring, single pulse energy, pulse interval, propulsion speed, and helical trajectory pitch, etc.

[0021] The beneficial effects of this invention are: This invention sets the detection light pulse energy, pulse width, and effective irradiation area. It irradiates the periapical region with low-energy detection light and collects photoacoustic signals using an ultrasonic micro-sensor array. The initial sound pressure amplitude and expected propagation time are estimated based on the absorbed energy density. The original time-domain waveforms of each sensor are then recorded and preprocessed, and gain correction and alignment are performed according to a time reference. The time-frequency energy distribution is obtained through short-time Fourier transform. Multiple spectral features are extracted within a sliding time window, compared with pre-calibrated dangerous spectral types, and risk scores are calculated. These are then compressed into a single risk value using Mahalanobis distance. When the value exceeds a threshold, it is determined to be near penetration or tissue stimulation. Subsequently, the microscopic morphology, elemental composition, and photoacoustic warning information are projected onto a unified three-dimensional grid. Multimodal alignment is achieved through coarse registration of feature points and fine registration using least-squares. The morphological features, elemental ratio vectors, and spectral features of each pixel are extracted, and the infection probability and confidence level of each modality are calculated. Weighted fusion is used to form an infection probability field. Isosurfaces are extracted through thresholding and topological repair and smoothing are performed to obtain... The three-dimensional cutting envelope of the primary tooth is analyzed, and preoperative and real-time images are uniformly registered with the cutting boundary to construct triangular and volumetric meshes. Element material parameters are generated through chemical composition mapping to establish a finite element simulation model. Subsequently, multiple candidate laser strategies are parameterized and the simulation tasks are executed in parallel within the model. Temperature field, cutting depth, and apical perturbation are calculated by constructing an energy injection function, and performance indicators are extracted for multi-objective scoring. Finally, the robustness of the strategies is evaluated through random perturbation and statistical analysis, and a set of priority laser execution strategies ranked by robustness is output. This approach can more accurately distinguish between infected and healthy dentin, reducing misjudgment or omission of infected areas, effectively reducing the risk of apical perforation and periapical tissue damage. The laser cutting process always revolves around the actual infected area, which can fully remove infected tissue while preserving healthy tooth structure to the greatest extent, improving the efficiency and safety of root canal preparation, reducing the doctor's reliance on experience, and making primary tooth root canal preparation operations more standardized and repeatable. Attached Figure Description

[0022] The present invention will now be further described with reference to the accompanying drawings.

[0023] Figure 1 This is a framework diagram of a laser-guided intelligent root canal preparation system for deciduous teeth. Detailed Implementation

[0024] 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.

[0025] 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.

[0026] Example 1: This embodiment of the invention provides a laser-guided intelligent root canal preparation system for deciduous teeth. See also... Figure 1 , Figure 1 This is a framework diagram of a laser-guided intelligent root canal preparation system for deciduous teeth, provided in an embodiment of the present invention. The system includes: an image processing module, a self-registration module, a structure acquisition module, a spectral processing module, an anomaly detection module, a fusion judgment module, a compensation tracking module, a digital modeling module, a simulation learning module, a trajectory generation module, an execution monitoring module, a safety protection module, an interactive command module, and a record auditing module.

[0027] The image processing module receives and preprocesses preoperative image data, performs preliminary segmentation of periapical structures, and generates a reference coordinate system.

[0028] Specifically, preoperative image data is read from imaging equipment or PACS, and the header information of each image is parsed to extract the pixel spacing, slice thickness, number of pixel rows and columns, and direction cosine matrix of each image. The image data are stacked into a three-dimensional array according to their pixel positions and slice spacing, and the original scanner coordinate system information is recorded. Then, a uniform target voxel size is determined based on clinical needs, and the corresponding position of each voxel in the original scanner coordinate system is calculated point by point on the target voxel grid. Trilinear interpolation is used to generate volume data after resampling of each image data, and the new pixel spacing and volume dimension are recorded. Local sub-blocks of a preset size are extracted centered on each resampled pixel position. Simultaneously, the similarity between each local sub-block and each sub-block within a preset search window is calculated, and corresponding weights are assigned based on the sub-block similarity. Then, a weighted average is applied to each local sub-block to replace the center pixel value to remove noise from the image data. Finally, low-pass filtering is used to calculate the pixel similarity. The image data is processed by calculating the logarithmic bias field and normalizing each pixel based on the acquired logarithmic bias field. The grayscale histogram of the corresponding image data is then calculated using the corrected pixels. Based on the grayscale distribution of the histogram, the image data is divided into candidate classes—background, dentin, and pulp—using the Otsu multi-threshold method. Three-dimensional connected component analysis is then performed on each candidate class, selecting connected components whose volume and shape indicators match the root canal characteristics. The pixel set of each connected component is centered, and its covariance matrix is ​​constructed. Principal component analysis is then used to determine the initial principal axis direction of the root canal, which is output as a point + direction vector. Simultaneously, the signal-to-noise ratio and contrast ratio of the ROI region in the image data are calculated, and the current image data quality is evaluated based on the calculation results. If the image data quality does not meet the preset standard, a prompt is made indicating whether the image data needs to be re-imported or the pixel size adjusted. The pixel spacing, volume dimension, coordinate transformation matrix, and initial root canal axis are recorded and output.

[0029] The self-test registration module is used to perform power-on self-tests, synchronous pulse timing, and delay correction on various hardware devices, and to establish a unified timestamp and trigger schedule. The structure acquisition module receives the registered hardware device signals and reference coordinate system, and performs high-frame-rate three-dimensional scanning of the root canal wall, outputting high-resolution microscopic morphological data.

[0030] Specifically, the sample arm and reference arm are installed according to the predetermined probe geometry. The initial estimate of the physical optical path difference between the sample and reference arms is recorded. Simultaneously, the original spectrum of the calibration white plate or reference mirror is acquired. The original spectrum is then smoothed and calibrated with instrument response correction to obtain the corrected reference spectrum. Based on the corrected reference spectrum, an index mapping table is constructed, and the original sampling points are resampled to equally spaced k points. Interpolation is then performed on each wavelength based on the same index mapping table to obtain the resampled spectral sequence. The DC component in each resampled spectral sequence is removed. Wavenumber-dependent compensation phase curves are calculated and saved as compensation phase terms by phase fitting for each wavelength. The obtained compensation phase terms are used to preprocess the processed spectral sequence. Then, a discrete inverse Fourier transform is performed on the preprocessed spectral sequence to obtain the complex-valued depth spectrum. Finally, envelope detection is performed on the complex-valued depth spectrum to obtain an amplitude-type A-scan curve, which is recorded as... The depth profile is calculated, and the time required for single-volume acquisition is matched with the acceptable range of motion in the child's oral cavity. After matching, multiple scans are performed in the lateral direction at preset intervals using a lateral scanning mechanism to form a set of two-dimensional lateral-depth planes. The acquisition of the lateral-depth planes is repeated, and three-dimensional volumes are constructed by acquiring data at different lateral angles. At the same time, A-scan curves are written into the volume buffer in frame form. Based on the cross-correlation method, adjacent lateral-depth planes in the child's oral cavity are compared to obtain corresponding displacement estimates. Then, the displacement estimates calculated for each frame are processed by inverse transformation in the volume buffer to compensate for structural displacement caused by motion. The three-dimensional volume obtained by discrete inverse Fourier transform is subjected to local contrast enhancement, multi-angle or multi-view composite speckle removal, and small-scale sharpening filtering to enhance the details of tubular openings and microcracks. After processing, the depth profile, surface point cloud, and local magnified view are exported.

[0031] The spectral processing module is used to capture plasma emission spectra and extract the characteristic information of each element from the raw emission spectrum based on the triggering time schedule; the anomaly detection module collects periapical photoacoustic signals, performs time-frequency analysis and spectral feature extraction, and identifies early warning spectral patterns of penetration or tissue stimulation in real time.

[0032] Specifically, the pulse energy and pulse width are selected, and the effective irradiation area of ​​the incident light in the local area is measured. Then, the periapical region is irradiated with low-energy short pulses or narrowband continuous wave probe light at a predetermined detection position using an ultrasonic microsensor. The local light flux and the initial light energy density absorbed by the tissue are calculated, and all parameters of each detection are recorded. Based on known physical relationships and the absorbed energy density, the initial ultrasonic pressure amplitude caused by local instantaneous absorption is calculated, and its spatial distribution estimate is recorded. The expected arrival time is calculated based on the geometric position of the sensor array and the distance between the generation point and the ultrasonic microsensor, and the expected arrival time of each ultrasonic microsensor is recorded as an initial alignment reference. Then, the original time-domain waveforms acquired by each ultrasonic microsensor are recorded with corresponding timestamps. Finally, the original time-domain waveforms are pre-amplified, the sampling rate is set, and AD is applied. C-quantization processing is performed, and anti-aliasing filtering is applied to each processed time-domain waveform. Then, based on the sensor bandwidth and target spectral range, the corresponding bandpass frequency band is selected and bandpass filtered. After that, based on the acquired initial alignment reference, normalized gain correction and time alignment are performed on the time-domain waveform of each sensor transmission channel. Through short-time Fourier transform, the energy distribution of each filtered and aligned time-domain waveform in the time-frequency plane is obtained. Based on a preset sliding time window, multiple sets of spectral features of each time-domain waveform in different time periods are extracted. Then, each set of spectral features is compared with the pre-calibrated corresponding hazard spectral type reference statistics to obtain the corresponding risk score. Then, the Mahalanobis distance function is used to compress the multidimensional features into a single risk value. If the risk value exceeds the set threshold, it is marked as near penetration or tissue stimulation within milliseconds, and the time window and corresponding features of this trigger are output.

[0033] The fusion judgment module receives and fuses microscopic morphological data, elemental feature information, and early warning spectrum information, and outputs the three-dimensional cutting envelope of the child's deciduous teeth in real time.

[0034] Specifically, the microscopic morphological data, elemental characteristic information, and early warning spectral information of each modality are projected onto a unified three-dimensional grid. Multiple sets of aligned feature points are used to perform coarse registration of each modality's data. Then, local least squares method is used to adjust each modality's data to a consistent pixel level, generating trimodal raw observations for each pixel position on a unified reference grid. Afterwards, morphological features, chemical element ratio vectors, and spectral features of each pixel on the three-dimensional grid are extracted. Logistic mapping is then used to obtain the infection probability of each modality, and the confidence value of each modality for each pixel is calculated. Finally, each modality is assigned a specific value. The confidence values ​​are normalized to form corresponding weighted coefficients. Based on the obtained weighted coefficients, the infection probabilities of each modality are fused into a unified infection probability score. Then, based on the infection probability score of each pixel, a corresponding fused probability field is generated. The fused probability field is a scalar field, and an initial isosurface is extracted in the pixel space according to a preset threshold. This is used as the infection boundary. The initial isosurface is then topologically repaired, and the surface of the repaired isosurface is obtained by smoothing iteratively using Laplacian. The three-dimensional cutting envelope of the child's deciduous teeth is obtained, and the corresponding boundary geometry and its corresponding local confidence are output.

[0035] Example 2: This embodiment of the invention provides a laser-guided intelligent root canal preparation system for deciduous teeth. See also... Figure 1 , Figure 1 This is a framework diagram of a laser-guided intelligent root canal preparation system for deciduous teeth, provided in an embodiment of the present invention. The system includes: an image processing module, a self-registration module, a structure acquisition module, a spectral processing module, an anomaly detection module, a fusion judgment module, a compensation tracking module, a digital modeling module, a simulation learning module, a trajectory generation module, an execution monitoring module, a safety protection module, an interactive command module, and a record auditing module.

[0036] The compensation tracking module receives the handheld device's pose, reference coordinate system, and image information in real time, and compensates for the child's jaw movements and corrects the handheld device's pose based on the information. The digital modeling module constructs a simulation model of the child's deciduous teeth based on preoperative image data, real-time image data, three-dimensional cutting envelope, and handheld device pose.

[0037] Specifically, pixel sets of preoperative image data and intraoperative real-time image data are collected and processed. Corresponding feature points are selected from both sets for initial estimation of rigid registration. This initial estimation is then refined through iterative minimization. The preoperative image data, real-time image data, and 3D cutting envelope are rigidly registered to the same reference system. The rigidly registered boundary surface is used as the outer surface and represented by a triangular mesh. Topological restoration is performed on the outer surface, generating a volume mesh inside the outer surface using a Delaunay-based volume subdivision method. A locally refined mesh is also generated at the root canal access points. The topology, geometry, and hierarchy of each mesh element are recorded, and a finite element mesh is generated. Element feature information is then interpolated into the mesh nodes or elements. Based on the chemical composition and structural indicators, a physical-empirical mapping rule is set, and based on the mapping rule, the chemical composition observation is converted into unit-level material parameters. At the same time, the unit-level material parameters are written into the finite element mesh. The corresponding physical model is selected according to clinical needs, and then the selected physical model is discretized on the finite element mesh to generate a simulation model of the child's deciduous teeth. The simulation model of the child's deciduous teeth is used to perform simulation prediction at a preset time step. At each sampling time, the simulation model of the child's deciduous teeth receives the pose of the handheld device and the latest image data. At the same time, the received data is mapped to the coordinate system of the child's deciduous teeth simulation model through registration transformation. Then, the latest data and the corresponding simulation prediction results are fused using a recursive state estimator, and the current field state of the child's deciduous teeth simulation model is corrected through incremental correction.

[0038] The simulation learning module runs multiple candidate execution strategies on a simulation model of the child's deciduous teeth and generates a list of strategy executions.

[0039] Specifically, each set of pre-defined candidate execution strategies is represented by corresponding parameterized variables, and each parameter is packaged into a column vector as a description of the corresponding candidate execution strategy. Then, a set of basis functions is selected to describe the cross-sectional shape of the optical cutting envelope, and a three-dimensional envelope approximation of the light field of any candidate light sheet is generated using a linear combination of basis functions. Based on available computing resources, each candidate execution strategy is mapped into parallel simulation tasks in batches, and the light energy injection function of each parallel task is instantiated in the simulation model of the child's deciduous teeth. Simulation time windows and step sizes are assigned to each task, and the parameters and simulation identifiers of each task are recorded. Then, based on the three-dimensional envelope approximation, the set pulse energy, and the repetition pattern, a spatiotemporal distribution of energy injection density is constructed for each parallel task. If the execution strategy is a pulse sequence, the instantaneous energy distribution corresponding to each pulse is superimposed in the time domain. If it is a continuous progression, it is mapped to a smooth time function. At the same time, the instantaneous injection during the simulation process is short-time averaged to generate the corresponding heat source term. In each parallel simulation, the instantaneous temperature field is solved by driving the heat conduction equation using the corresponding heat source term. Based on the instantaneous temperature field and energy absorption, a corresponding cutting depth field is established. Then, the equivalent nodal force generated by plasma expansion or impact is used as an external load, and the displacement response of the root tip and adjacent tissues is calculated to obtain the root tip perturbation metric. After each simulation task, various indicators are extracted from the solution field, and the obtained indicators are normalized and mapped to corresponding performance vectors. Each performance vector is scored according to the multi-objective aggregation criterion, and a candidate strategy list and a metric detail for each strategy are output based on the score, the magnitude of the score, and a preset quantity threshold. Then, the parameters of each execution strategy in the candidate strategy list are randomly perturbed, and the model simulation is repeated to generate the corresponding score statistical distribution. The expected score and score standard deviation of each candidate strategy are calculated, and the candidates are re-ranked based on the preset robustness index, outputting a set of preferred strategies ranked by robustness.

[0040] The trajectory generation module calculates the optical phase mask and outputs the successively triggered optical settings and scanning trajectory based on the strategy execution list and the three-dimensional cutting envelope. The execution monitoring module drives the femtosecond laser to perform cutting based on the generated scanning trajectory and collects various data in real time for online analysis.

[0041] The safety protection module receives real-time spectral data, image data, simulation data, and analysis results, performs online risk assessment, selects to trigger multimodal warnings, records events, and rolls back to safety parameters; the interactive command module displays various data during the operation, while the operator manually selects or overrides suggestions and records signature confirmation; the record audit module summarizes various data of this operation and writes them into the experience database, while generating postoperative auditable reports and review data.

[0042] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A laser-guided intelligent root canal preparation system for deciduous teeth, characterized in that, include: The system includes an image processing module, a self-registration module, a structure acquisition module, a spectral processing module, an anomaly detection module, a fusion judgment module, a compensation tracking module, a digital modeling module, a simulation learning module, a trajectory generation module, an execution monitoring module, a security protection module, an interactive command module, and a record auditing module. The image processing module receives and preprocesses preoperative image data, initially segments the periapical structure, and generates a reference coordinate system. The self-test registration module is used to perform power-on self-tests, synchronization pulse timing, and delay corrections on various hardware devices, and to establish a unified timestamp and trigger schedule. The structure acquisition module receives the registered hardware device signal and reference coordinate system, and performs high-frame-rate three-dimensional scanning of the root canal wall, outputting high-resolution microscopic morphological data. The spectral processing module is used to capture plasma emission spectra and extract elemental characteristic information from the raw emission spectra based on a triggering schedule. The anomaly detection module collects periapical photoacoustic signals, performs time-frequency analysis and spectral feature extraction, and identifies early warning spectral patterns of penetration or tissue stimulation in real time. The fusion judgment module receives and fuses microscopic morphological data, elemental feature information, and early warning spectrum information, and outputs the three-dimensional cutting envelope of the child's deciduous teeth in real time. The compensation tracking module receives the handheld device's pose, reference coordinate system, and image information in real time, and compensates for the child's jaw movements and corrects the handheld device's pose based on the information. The digital modeling module constructs a simulation model of the child's deciduous teeth based on preoperative image data, real-time image data, three-dimensional cutting envelope, and handheld device pose. The simulation learning module runs multiple candidate execution strategies on the child's deciduous teeth simulation model and generates a strategy execution list; The trajectory generation module calculates the optical phase mask and outputs the successively triggered optical settings and scanning trajectory based on the strategy execution list and the three-dimensional cutting envelope. The execution monitoring module drives the femtosecond laser to perform cutting based on the generated scanning trajectory, and collects various data in real time for online analysis; The security protection module receives real-time spectral data, image data, simulation data, and analysis results, performs online risk assessment, selects to trigger multimodal warnings, records events, and rolls back to security parameters. The interactive command module is used to display various data during the operation, while the operator can manually select or overwrite suggestions and record signature confirmation. The record audit module is used to summarize all data from this surgery and write them into the experience database, while generating an auditable postoperative report and review data.

2. The laser-guided intelligent root canal preparation system for deciduous teeth according to claim 1, characterized in that, The specific steps of the image processing module in initially segmenting the periapical structure and generating a reference coordinate system are as follows: S1.1: Read preoperative image data from imaging equipment or PACS, then parse the header information of each image data and extract the pixel spacing, slice thickness, number of pixel rows and columns and direction cosine matrix of each image data. Stack each image data into a three-dimensional array according to its pixel position and slice spacing, and record the original scanner coordinate system information. Then, determine the uniform target voxel size according to clinical needs, calculate the corresponding position of each point on the target voxel grid in the original scanner coordinate system, and generate the volume data after resampling of each image data through trilinear interpolation, and record the new pixel spacing and volume dimension. S1.2: Extract local sub-blocks of a preset size centered on each pixel position after each resampling. Simultaneously calculate the similarity between each local sub-block and each sub-block within the preset search window. Assign corresponding weights based on the sub-block similarity. Then, replace the center pixel value by performing a weighted average on each local sub-block to remove noise from each image data. Next, use low-pass filtering to calculate the logarithmic bias field of each pixel and perform normalization correction on each pixel based on the obtained logarithmic bias field. S1.3: Calculate the grayscale histogram of the corresponding image data using each corrected pixel. Based on the grayscale distribution of the grayscale histogram, divide the image data into candidate classes of background, dentin, and pulp using the Otsu multi-threshold method. Then, perform three-dimensional connected component analysis on each candidate class and select connected components whose volume and shape indicators conform to the root canal characteristics. S1.4: After centering the pixel set of each connected component, construct its covariance matrix. Then, determine the initial principal axis direction of the root canal through principal component analysis and output the initial principal axis direction of the root canal in the form of point + direction vector. At the same time, calculate the signal-to-noise ratio and contrast index of the ROI region in the image data, and evaluate the current image data quality based on the calculation results. If the image data quality does not meet the preset standard, prompt whether it is necessary to re-import the image data or adjust the pixel size. Record and output the pixel spacing, volume dimension, coordinate system transformation matrix and initial root canal axis.

3. The laser-guided intelligent root canal preparation system for deciduous teeth according to claim 1, characterized in that, The specific steps by which the structure acquisition module performs high-frame-rate three-dimensional scanning of the root canal wall and outputs high-resolution microscopic morphological data are as follows: S2.1: Install the sample arm and reference arm according to the predetermined probe geometry, record the initial estimate of the physical optical path difference between the sample and the reference arm, and simultaneously acquire the original spectrum of the calibration white plate or reference mirror. Then, smooth and correct the instrument response of the original spectrum to obtain the corrected reference spectrum. Based on the corrected reference spectrum, construct a set of index mapping tables, and resample the original sampling points to k points at equal intervals. Then, interpolate each wavelength based on the same index mapping table to obtain the resampled spectral sequence. S2.2; Remove the DC component from each resampled spectral sequence, and calculate the wavenumber-dependent compensation phase curve by phase fitting for each wavelength and save it as a compensation phase term. Then, perform phase preprocessing on the processed spectral sequence using the calculated compensation phase term, and then perform discrete inverse Fourier transform on the preprocessed spectral sequence to obtain the complex-valued depth spectrum. After that, perform envelope detection on the complex-valued depth spectrum to obtain the amplitude-type A-scan curve and record it as a depth profile. S2.3: Calculate the time required for single volume acquisition and match it with the acceptable range of motion in the child's oral cavity. After matching, use the lateral scanning mechanism to perform multiple scans in the lateral direction at preset intervals to form a set of two-dimensional lateral-depth planes. Repeat the acquisition of the lateral-depth planes and acquire three-dimensional volumes at different lateral angles. At the same time, write the A-scan curve into the volume buffer in frame form. S2.4: Based on the cross-correlation method, the adjacent transverse-depth planes in the child's oral cavity are compared to obtain the corresponding displacement estimates. Then, the displacement estimates calculated for each frame are inversely transformed in the volume buffer to compensate for the structural displacement caused by the motion. S2.5: The 3D volume obtained by discrete inverse Fourier transform is subjected to local contrast enhancement, multi-angle or multi-view composite despeccization, and small-scale sharpening filtering to enhance the details of the tube opening and microcracks. After processing, the depth section, surface point cloud, and local magnified view are exported.

4. The laser-guided intelligent root canal preparation system for deciduous teeth according to claim 1, characterized in that, The specific steps of the anomaly detection module in real-time identifying the warning spectrum of penetration or tissue stimulation are as follows: S3.1: Select the pulse energy and pulse width, and measure the effective irradiation area of ​​the incident light in the local area. Then, use a low-energy short pulse or narrow-band continuous wave probe light to irradiate the periapical region at the predetermined detection position using an ultrasonic microsensor. Calculate the local light flux and the initial light energy density absorbed by the tissue. Record all parameters for each detection. Based on the known physical relationship and the absorbed energy density, calculate the initial ultrasonic pressure amplitude caused by local instantaneous absorption and record its spatial distribution estimate. S3.2: Calculate the expected arrival time based on the geometric position of the sensor array and the distance between the generation point and the ultrasonic microsensor, and record the expected arrival time of each ultrasonic microsensor as an initial alignment reference. Then, record the original time-domain waveforms acquired by each ultrasonic microsensor and attach the corresponding timestamps. Finally, perform pre-amplification, sampling rate setting and ADC quantization processing on each original time-domain waveform. S3.3: Perform anti-aliasing filtering on each processed time-domain waveform, then select the corresponding bandpass frequency band based on the sensor bandwidth and target spectrum range, and perform bandpass filtering on it. After that, based on the obtained initial alignment reference, perform normalized gain correction and time alignment on the time-domain waveform of each sensor transmission channel. S3.4: By using short-time Fourier transform, the energy distribution of each filtered and aligned time-domain waveform in the time-frequency plane is obtained. Based on a preset sliding time window, multiple sets of spectral features of each time-domain waveform in different time periods are extracted. Then, each set of spectral features is compared with the pre-calibrated corresponding hazard spectrum reference statistics to obtain the corresponding risk score. Then, the Mahalanobis distance function is used to compress the multidimensional features into a single risk value. If the risk value exceeds the set threshold, it is marked as close to penetration or tissue stimulation within milliseconds, and the time window and corresponding features of this trigger are output.

5. The laser-guided intelligent root canal preparation system for deciduous teeth according to claim 4, characterized in that, The specific steps of the fusion determination module in real-time outputting the three-dimensional cutting envelope surface of the child's deciduous teeth are as follows: S4.1: Project the microscopic morphological data, elemental characteristic information, and early warning spectrum information of each modality onto a unified three-dimensional grid. Use multiple sets of aligned feature points to perform coarse registration on each modality data. Then, use the local least squares method to adjust each modality data to a consistent pixel level to generate the three-modal raw observations at each pixel position on the unified reference grid. After that, extract the morphological features, chemical element ratio vectors, and spectral features of each pixel on the three-dimensional grid. And obtain the infection probability of each modality through Logistic mapping. S4.2: Calculate the modal confidence value for each pixel, normalize each modal confidence value to form a corresponding weighting coefficient, and based on the obtained weighting coefficient, fuse the infection probabilities of each modality into a unified infection probability score. Then, generate the corresponding fused probability field based on the infection probability score of each pixel. S4.3: The fusion probability field is converted into a scalar field, and an initial isosurface is extracted in the pixel space according to a preset threshold. This is then used as the infection boundary. The initial isosurface is then topologically repaired, and the surface of the repaired isosurface is smoothed by Laplacian smoothing to obtain the three-dimensional cutting envelope of the child's deciduous teeth. At the same time, the geometry of the corresponding boundary surface and its corresponding local confidence are output.

6. The laser-guided intelligent root canal preparation system for deciduous teeth according to claim 5, characterized in that, The specific steps for constructing a simulation model of the child's deciduous teeth using the digital modeling module are as follows: S5.1: Collect the pixel data set of the preoperative image data and the pixel data set of the intraoperative real-time image data, and select the corresponding feature points from the two sets of pixel data to perform the initial estimation of rigid registration. Then, perform fine registration on the initial estimation by iterative minimization, and transform the preoperative image data, real-time image data and three-dimensional cutting envelope into the same reference system through rigid registration. S5.2: The rigidly registered boundary surface is used as the outer surface and represented in the form of a triangular mesh. At the same time, the topology of the outer surface is repaired. A volume mesh is generated inside the outer surface using a Delaunay-based volume meshing method. Meanwhile, a locally refined mesh is generated for the root canal passage. Then, the topology, geometry and hierarchy of each mesh element are recorded, and a finite element mesh is generated. S5.3: Interpolate elemental feature information to grid nodes or cell centers, set physical-empirical mapping rules based on chemical composition and structural indicators, and convert chemical composition observations into cell-level material parameters based on the mapping rules. Simultaneously, write the cell-level material parameters into the finite element mesh, select the corresponding physical model according to clinical needs, and then perform finite element discretization on the finite element mesh to generate a simulation model of the child's deciduous teeth. Simulation prediction is then performed on the child's deciduous teeth simulation model at a preset time step. S5.4: The child's deciduous teeth simulation model receives the handheld device pose and the latest image data at each sampling time. At the same time, it maps the received data into the coordinate system of the child's deciduous teeth simulation model through registration transformation. Then, it uses a recursive state estimator to fuse the latest data with the corresponding simulation prediction results and corrects the current field state of the child's deciduous teeth simulation model through incremental correction.

7. The laser-guided intelligent root canal preparation system for deciduous teeth according to claim 6, characterized in that, The specific steps of the simulation learning module in running multiple candidate execution strategies on the child's deciduous tooth simulation model and generating a strategy execution list are as follows: S6.1: Each set of preset candidate execution strategies is represented by a corresponding parameterized variable, and each parameter is packaged into a set of column vectors as a description of the corresponding candidate execution strategy. Then, a set of basis functions is selected to describe the cross-sectional shape of the optical cutting envelope, and the three-dimensional envelope approximation of the optical field of any candidate light sheet is generated by linear combination of basis functions. S6.2: Based on available computing resources, each candidate execution strategy is mapped to a parallel simulation task in batches, and the light energy injection function of each parallel task is instantiated in the simulation model of the child's deciduous teeth. Simulation time windows and step sizes are assigned to each task, and the parameters and simulation identifiers of each task are recorded. Then, based on the three-dimensional envelope approximation, the set pulse energy and repetition pattern, the spatiotemporal distribution of energy injection density is constructed for each parallel task. S6.3: If the execution strategy is a pulse sequence, the instantaneous energy distribution corresponding to each pulse is superimposed in the time domain. If it is a continuous progression, it is mapped to a smooth time function. At the same time, the instantaneous injection in the simulation process is averaged in a short time to generate the corresponding heat source term. Then, in each parallel simulation, the corresponding heat source term is used to drive the heat conduction equation to solve the instantaneous temperature field. S6.4: Based on the instantaneous temperature field and energy absorption, a corresponding cutting depth field is established. Then, the equivalent nodal force generated by plasma expansion or impact is used as an external load, and the displacement response of the root tip and adjacent tissues is calculated to obtain the root tip disturbance measurement. After each simulation task is completed, various indicators are extracted from the solution field. S6.5: The acquired indicators are normalized and mapped to corresponding performance vectors. Each performance vector is scored according to the multi-objective aggregation criterion. Based on the score and a preset quantity threshold, a candidate strategy list and the measurement details of each strategy are output. Then, the parameters of each execution strategy in the candidate strategy list are randomly perturbed, and the model simulation is repeated to generate the corresponding score statistical distribution. The expected score and score standard deviation of each candidate strategy are calculated. Based on the preset robustness index, the candidates are reordered, and the set of priority strategies ordered by robustness is output.

Images