Clinical effect evaluation system for constant pulpitis live pulp preservation treatment scheme

Abstract

The invention relates to the field of precision diagnosis and treatment evaluation of oral medicine and artificial intelligence crossing in biomedical engineering, in particular to a clinical effect evaluation system of a permanent tooth pulpitis live pulp preservation treatment scheme, which comprises a preoperative evaluation module used for acquiring a preoperative physiological index set and processing the preoperative physiological index set to obtain an evaluation result of the preoperative physiological index set; the preoperative state of pulpitis of permanent teeth is obtained, so that whether a live pulp preservation treatment scheme is implemented for the permanent teeth or not is determined; the preoperative physiological index set comprises MPI, TOI, CMII and MSI; the preoperative state comprises an inflammation state and an activity state of permanent teeth; and the postoperative effect evaluation module is used for acquiring a postoperative physiological index set in at least one time period after the live pulp preservation treatment scheme is finished, and comprehensively evaluating the postoperative physiological index set to obtain the clinical CES of pulpitis. According to the scheme, a full-cycle accurate quantification technology for pulpitis treatment is provided.

Description

Technical Field

[0001] This application relates to the field of precision diagnosis and evaluation at the intersection of oral medicine and artificial intelligence in biomedical engineering, and includes, but is not limited to, a clinical efficacy evaluation system for vital pulp preservation treatment plans for permanent tooth pulpitis. Background Technology

[0002] Accurate diagnosis, targeted treatment, and follow-up evaluation of treatment effects for dental pulp diseases, including pulpitis, can directly impact patients' short-term lifestyle, daily diet, and long-term oral health.

[0003] In practice, before treating pulpitis, the severity of the pulpitis is usually assessed based on the dentist's experience or single-dimensional data to determine whether a vital pulp preservation treatment plan should be implemented. After treatment, the effectiveness of the vital pulp preservation treatment plan is usually evaluated using scattered and isolated data. However, the above approach lacks comprehensiveness and cannot achieve closed-loop clinical management of pulpitis from diagnosis and treatment to postoperative follow-up. Therefore, there is an urgent need for quantitative techniques covering the entire treatment cycle of pulpitis to support clinical decision-making. Summary of the Invention

[0004] To address the aforementioned problems in the prior art, this application provides a clinical efficacy evaluation system for a vital pulp preservation treatment plan for permanent tooth pulpitis. This system can achieve precise quantification of the entire treatment cycle for pulpitis, thereby providing support for clinical decision-making.

[0005] The specific technical solution of this application embodiment is as follows: This application provides a clinical efficacy evaluation system for a permanent tooth pulpitis vital pulp preservation treatment plan, including: The preoperative assessment module is used to acquire a set of preoperative physiological indicators and process the set of preoperative physiological indicators to obtain the preoperative status of pulpitis in the permanent tooth, so as to determine whether the pulp preservation treatment plan should be implemented for the permanent tooth. The set of preoperative physiological indicators includes the Microcirculatory Perfusion Index (MPI), Tissue Oxygenation Index (TOI), Cell Membrane Integrity Index (CMII), and Metabolic Stress Index (MSI). The preoperative status includes the inflammatory status and vitality status of the permanent tooth. The postoperative efficacy evaluation module is used to acquire a set of postoperative physiological indicators at least once after the completion of the pulp preservation treatment plan, and to comprehensively evaluate the set of postoperative physiological indicators to obtain the Comprehensive Efficacy Score (CES) for pulpitis. The set of postoperative physiological indicators includes the Functional Survival Index (FSI), the Structural Repair Index (SRI), and the Symptomatic Improvement Index (SII).

[0006] The beneficial effects of the technical solutions provided in this application include at least the following: The clinical efficacy evaluation system for the vital pulp preservation treatment plan for permanent teeth pulpitis provided in this application includes a preoperative assessment module. This module acquires a set of preoperative physiological indicators, including MPI, TOI, CMII, and MSI, and processes these indicators to obtain the preoperative state of pulpitis in the permanent tooth. This allows for comprehensive collection of physiological indicators of the patient's permanent tooth in a pulpitis state before the implementation of the vital pulp preservation treatment plan, thereby improving the completeness and accuracy of the preoperative state. Furthermore, the preoperative state includes both the inflammatory state and the vitality state of the permanent tooth, enabling accurate assessment of the severity of pulpitis from both inflammation and vitality dimensions. Based on this, the system determines whether to implement a vital pulp preservation treatment plan for the permanent tooth, thus enabling the determination of whether to proceed. The preoperative assessment module allows for precise judgment of pulp preservation treatment plans. On the other hand, the postoperative efficacy evaluation module acquires a set of postoperative physiological indicators (FSI, SRI, and SII) for at least one period after the completion of the pulp preservation treatment plan, thus improving the comprehensiveness of the indicators. Furthermore, a comprehensive evaluation of the postoperative physiological indicators yields the Clinical Comprehensive Evaluative Score (CSE) for pulpitis, improving the accuracy of the CSE. Moreover, the organic combination of the preoperative assessment module and the postoperative efficacy evaluation module enables precise, multi-stage, and comprehensive clinical closed-loop management from the diagnosis and treatment of pulpitis to postoperative follow-up, providing a quantitative technology for the entire treatment cycle of pulpitis and thus offering strong support for clinical decision-making in pulpitis treatment. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in the embodiments of this application; Figure 2Another schematic diagram of the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the dual-task regression neural network provided in an embodiment of this application. Detailed Implementation

[0008] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0009] Accurate diagnosis, targeted treatment, and follow-up evaluation of treatment effects for dental pulp diseases, including pulpitis, can directly impact a patient's short-term lifestyle, daily diet, and long-term oral health.

[0010] In practice, before treating pulpitis, the severity of pulpitis is usually assessed based on the doctor's clinical experience or by using single-dimensional data to determine whether to perform vital pulp preservation. For example, the determination of whether to perform vital pulp preservation is based solely on electrical vitality test data before treatment, which lacks systematic quantification of multiple physiological parameters.

[0011] Meanwhile, after treatment, the results of viable bone marrow preservation are usually evaluated using scattered and isolated data. However, patient follow-up data is scattered and isolated, and the standardization of the aforementioned data preprocessing and fusion methods is also low.

[0012] On the other hand, the entire process from diagnosis to treatment and even the end of treatment lacks cross-stage collaboration, making it difficult to achieve accurate and dynamic tracking from diagnosis to follow-up, and unable to achieve closed-loop clinical management of pulpitis from diagnosis to postoperative follow-up. Therefore, there is an urgent need for a quantitative technology for the entire cycle of pulpitis treatment to support clinical decision-making.

[0013] Based on the above technical problems, this application provides a clinical efficacy evaluation system for a permanent tooth pulpitis vital pulp preservation treatment plan. Figure 1 This is a schematic diagram of the structure of the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in the embodiments of this application, as shown below. Figure 1 As shown, the clinical efficacy evaluation system 100 may include: The preoperative assessment module 101 is used to acquire a set of preoperative physiological indicators and process the set of preoperative physiological indicators to obtain the preoperative status of pulpitis in permanent teeth, so as to determine whether to implement a vital pulp preservation treatment plan for permanent teeth. The set of preoperative physiological indicators includes MPI, TOI, CMII and MSI; the preoperative status includes the inflammatory status and vitality status of permanent teeth.

[0014] The postoperative effect evaluation module 102 is used to obtain a set of postoperative physiological indicators at least once after the end of the pulp preservation treatment plan, and to comprehensively evaluate the set of postoperative physiological indicators to obtain the clinical CES of pulpitis; wherein, the set of postoperative physiological indicators includes FSI, SRI and SII.

[0015] In some embodiments, the set of preoperative physiological indicators can be obtained in the following ways: By acquiring pulp-related physiological signals through acquisition equipment and preprocessing the pulp-related physiological signals to obtain preprocessing results, and then calculating physiological indicators from the preprocessing results, various preoperative physiological indicators in the preoperative physiological indicator set can be obtained; for example, preprocessing may include noise reduction processing, completion and calibration processing, and normalization processing.

[0016] Specifically, dental pulp-related physiological signals can be obtained in the following ways: A laser Doppler flowmeter (LDF) is used to acquire blood perfusion data and pulsation waveform data of the pulp tissue of permanent teeth with pulpitis. Specifically, the LDF continuously acquires blood perfusion signals of the pulp tissue over a period of time and converts them into continuous perfusion unit (PU) values. The arithmetic mean of these continuous PU values ​​is then taken to obtain the average PU value of the pulp blood perfusion. Next, the standard deviation of the PU value is calculated based on the average PU value. At the same time, the pulsation index is calculated by dividing the difference between the peak and trough values ​​of the acquired pulsation waveform data by the mean. Thus, the pulp-related physiological signals can include blood perfusion data and pulsation waveform data, and the acquisition device can include an LDF.

[0017] The intensity changes of red and infrared light after passing through tissues and fluctuating with the pulse are acquired using a pulse oximeter. Then, using the Lambert-Beer law, the light absorption contribution of arterial blood is separated, and the absorption ratio of oxyhemoglobin to deoxyhemoglobin at the two wavelengths is calculated to obtain arterial oxygen saturation (SO2). In this case, dental pulp-related physiological signals may include SO2, and the acquisition device may include a pulse oximeter.

[0018] An electrical impedance spectrometer (EIS) is used to apply alternating current (AC) signals of different frequencies to the dental pulp tissue, measuring its impedance. Simultaneously, a low-frequency AC signal of a specific frequency is applied to detect the impedance response of the pulp tissue, acquiring raw impedance data containing multiple frequency points. This lays the foundation for subsequent extraction of cell membrane capacitance (Cm) and extracellular fluid resistance (Re). In this case, the dental pulp-related physiological signals can include raw impedance data at multiple frequency points, and the acquisition device can include an EIS.

[0019] The concentrations of oxygenated hemoglobin (HbO2) and deoxygenated hemoglobin (Hb) in the target tissue region are acquired using a near-infrared spectroscopy method. Then, the total hemoglobin (HbT) is calculated by summing HbO2 and Hb. Finally, the HbO2 concentration is divided by the HbT concentration to obtain the HbO2 to HbT ratio. In this case, the pulp-related physiological signals can include the concentrations of HbO2 and Hb, and the acquisition equipment can include a near-infrared spectroscopy device.

[0020] Specifically, data preprocessing may include the following operations: First, noise reduction is performed. For example, for LDF and blood oxygen data, Kalman filtering can be used to remove artifacts caused by gingival movement of permanent teeth and retain the real blood flow pulsation characteristics. For light intensity change signals, an adaptive iterative reweighted penalized least squares algorithm can be used to eliminate baseline drift caused by enamel scattering of permanent teeth. For impedance values, high-frequency noise can be eliminated by smoothing with a 5th-order polynomial.

[0021] Secondly, there is the need for completion and calibration. For missing data accounting for less than 5%, the K-Nearest Neighbors (KNN) interpolation method (K value set to 5) based on the same tooth position and age group can be used for completion. The impedance value can be fitted by the Cole-Cole model to complete the abnormal points in the frequency scan and extract core physiological parameters such as Cm and extracellular fluid resistance (Re).

[0022] Finally, normalization is performed, which maps all the above data to the [0,1] interval using the Min-Max method to eliminate the dimensional differences between data from different devices.

[0023] Specifically, the core objective of physiological index calculation can be defined as: transforming preprocessing results into intermediate indicators with clear biological significance, in order to reduce data dimensionality and enhance interpretability.

[0024] For MPI, it can be calculated based on the average PU value and pulsatility index collected by LDF. For example, the normalized average PU value and pulsatility index can be weighted and summed based on the first weight to obtain the first result. Then, the first result is multiplied by 100 to convert the first result into MPI with a value range of [0, 100]. The first weight can be (0.6, 0.4).

[0025] For TOI, it integrates the ratio of HbO2 to HbT measured by near-infrared spectroscopy and SO2. In the calculation, the normalized values ​​of the above two parameters are weighted and summed with a second weight to obtain a second result. Then, the second result is multiplied by 100 to obtain the TOI with a value range of [0, 100]. The second weight can include (0.7, 0.3). TOI quantifies the oxygen supply and consumption balance of dental pulp tissue. If TOI is greater than 80, it can indicate that the dental pulp tissue has sufficient oxygen supply, corresponding to a healthy dental pulp state. If TOI is less than 50, it can indicate that the dental pulp tissue is hypoxic, and there may be a risk of inflammation or necrosis.

[0026] For CMII, multiplying the normalized Cm value by 100 yields the CMII value, which ranges from [0, 100]. In practical applications, since Cm is directly related to the total area and integrity of the cell membrane, a CMII greater than 80 indicates that the cell membrane is intact and the dental pulp cells are in good condition, while a CMII less than 30 indicates extensive cell necrosis and severe damage to cell membrane function.

[0027] For MSI, a quantitative index can be calculated by fusing three core dental pulp physiological parameters using a fuzzy logic algorithm, as detailed below: First, the MPI coefficient of variation, TOI, and Re were obtained. Then, the above data were divided into three fuzzy sets: high, medium, and low, based on clinical physiological thresholds. Among them, the MPI coefficient of variation includes the ratio of the standard deviation of pulp blood flow perfusion to the mean PU value, which is used to reflect the stability of blood flow fluctuations.

[0028] Next, inference rules are pre-set based on the pathophysiological mechanism of dental pulp. For example, when a combination of parameters such as high MPI coefficient of variation (intense blood flow fluctuations), low TOI (tissue hypoxia), and high Re (extracellular fluid abnormality / cell damage) occurs, a high-stress inference conclusion can be triggered. This rule corresponds to the physiological characteristics of dental pulp in an acute inflammatory state.

[0029] Finally, the centroid method was used to defuzzify the low, medium, and high stress fuzzy conclusions derived from the reasoning, transforming the abstract fuzzy judgments into MSI values ​​ranging from [0, 100]. From a biological perspective, the core function of MSI is to assess the degree of inflammatory stress in dental pulp tissue. An MSI greater than 70 indicates a severe inflammatory response in the dental pulp, while an MSI less than 30 indicates a stable metabolic state of the dental pulp with no obvious inflammatory stress. If the value is between 30 and 70, it corresponds to moderate stress or an early inflammatory state, requiring further judgment in conjunction with other indicators.

[0030] In some implementation sets, the inflammatory status can characterize the severity of inflammation in permanent teeth and may also include the distribution of inflammation; correspondingly, the vitality status can characterize the pulp physiological function and survival potential of permanent teeth. The inflammatory status and vitality status can be represented by text data or numerical data, respectively.

[0031] In some embodiments, the decision to implement a pulp preservation treatment for permanent teeth can be determined in the following ways: If the inflammation status is greater than or equal to the first threshold and the vitality status is less than or equal to the second threshold, the bone marrow preservation treatment plan may not be implemented; if the inflammation status is less than the first threshold and the vitality status is greater than the second threshold, the bone marrow preservation treatment plan may be implemented. The first and second thresholds can be adjusted according to the patient's age and / or physical condition, and this application does not limit this adjustment.

[0032] In some embodiments, the length and number of at least one time period can be adjusted according to the patient’s age and / or physical condition.

[0033] In some embodiments, the postoperative physiological index set can accurately characterize the effect of the pulp preservation treatment plan from three core dimensions: pulp function survival status, structural repair quality, and clinical symptom improvement status.

[0034] Specifically, for FSI, patients can be followed up for at least one time period, and microcirculation perfusion index (MPI-s) and tissue oxygenation index (TOI-s) can be collected. The above indices are then weighted and summed according to a third weight to obtain a third result. The third result is then converted into FSI with a value range of [0, 100]. The third weight may include (0.6, 0.4).

[0035] For SRI, the semantic segmentation model U-Net can be used to automatically identify the dentin bridge region in cone-beam computed tomography (CBCT) images. The ratio of dentin bridge thickness to pulp chamber floor diameter and the ratio of dentin bridge density to normal dentin density are calculated. These two ratios are then weighted and integrated according to a fourth weight to obtain a fourth result. Subsequently, the Canny edge detection algorithm is used to identify the marginal gap between the filling and the tooth based on the fourth result. If the maximum marginal gap is greater than 50 μm, the fourth result is subtracted by 20; if the maximum gap is greater than 100 μm, the fourth result is subtracted by 5. The final output SRI has a value range of [0, 100] to evaluate the structural repair integrity of the dental pulp tissue. The fourth weight can include (0.5, 0.5).

[0036] For SII, it can be calculated using equation (1): (1) in, Visual Analog Scale (VAS) score representing preoperative pain prior to implementation of a viable bone marrow preservation treatment regimen. The VAS score representing postoperative pain at least one follow-up period. This represents the percussion pain level of the patient's permanent teeth during at least one follow-up period. Representing malocclusion, SII can quantify the degree of relief of a patient's subjective symptoms within the range of [0, 100].

[0037] In some embodiments, a comprehensive evaluation of the postoperative physiological indicators to obtain CES can be achieved in the following ways: The severity of the condition is determined by the inflammatory and vitality states in the preoperative state. The postoperative recovery status is determined based on the physiological indicators in the postoperative physiological indicator set. If the recovery status indicates that the patient's permanent teeth have relieved symptoms after surgery, it is determined as the first score. If the recovery status indicates that the patient's permanent teeth have relieved symptoms after surgery, it is determined as the second score. The first score can be greater than the second score.

[0038] As can be seen from the above, in the clinical efficacy evaluation system for the vital pulp preservation treatment plan for permanent teeth provided in this application embodiment, the preoperative evaluation module is used to acquire a set of preoperative physiological indicators, including MPI, TOI, CMII, and MSI, and to process the set of preoperative physiological indicators to obtain the preoperative state of pulpitis in the permanent tooth. Thus, before the implementation of the vital pulp preservation treatment plan, comprehensive collection of physiological indicators of the patient's permanent tooth in a pulpitis state is achieved, thereby improving the completeness and accuracy of the preoperative state. Furthermore, the preoperative state includes both the inflammatory state and the vitality state of the permanent tooth, thus enabling accurate assessment of the severity of pulpitis from both inflammation and vitality dimensions. Based on this, determining whether to implement a vital pulp preservation treatment plan for the permanent tooth based on the preoperative state allows for precise assessment of whether... The preoperative assessment module allows for precise determination of whether or not to implement a vital pulp preservation treatment plan. On the other hand, the postoperative efficacy evaluation module is used to obtain a set of postoperative physiological indicators at least for a period of time after the completion of the vital pulp preservation treatment plan. This set of indicators includes FSI, SRI, and SII, thus improving the comprehensiveness of the indicators in the postoperative physiological indicator set. Furthermore, a comprehensive evaluation of the postoperative physiological indicator set yields the clinical comprehensive score (CSE) for pulpitis, improving the accuracy of the CSE. Moreover, through the organic combination of the preoperative assessment module and the postoperative efficacy evaluation module, a comprehensive, multi-stage clinical closed-loop precise management system can be achieved from the diagnosis and treatment of pulpitis to postoperative follow-up, providing a quantitative technology for the entire treatment cycle of pulpitis and thus providing strong support for clinical decision-making in pulpitis.

[0039] Figure 2 Another schematic diagram of the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in the embodiments of this application is shown below. Figure 2 As shown: The preoperative assessment module 101 includes at least a dual-task regression neural network 103; wherein: The dual-task regression neural network 103 includes at least a shared feature layer 104 and a feature attention layer 105; wherein: The shared feature layer 104 is used to extract features from the preoperative physiological index set to obtain a high-order feature representation set of the preoperative physiological index set; the feature attention layer 105 is used to remove redundant data in the high-order feature representation set to obtain effective preoperative data for pulpitis.

[0040] Among them, the effective preoperative data is correlated with the preoperative condition.

[0041] In some embodiments, the dual-task regression neural network 103 can perform feature extraction, identification, and classification of a set of preoperative physiological indicators.

[0042] In some embodiments, the higher-order feature representation set may include the correlations and regularities among data in the preoperative physiological indicator set, which can comprehensively characterize the severity of pulpitis in a global and semantic manner.

[0043] In some embodiments, the shared feature layer may include at least two feature extraction units, and adjacent feature extraction units may be connected by fully connected layers to achieve a stacked configuration of at least two feature extraction units; correspondingly, the set of higher-order feature representations may be obtained in the following manner: The first feature extraction unit in the shared feature layer extracts features from the preoperative physiological index set to obtain the first extraction result. Then, the second feature extraction unit in the shared feature layer extracts features from the first extraction result, and so on, until the last feature extraction unit in the shared feature layer extracts features from the extraction result output by the second-to-last feature extraction unit, thereby obtaining a set of higher-order feature representations.

[0044] In some embodiments, redundant data may include interference data in the set of higher-order feature representations that are not necessarily related to pulpitis, and may also include noise data generated by the acquisition device during the acquisition of pulp-related physiological signals and the preprocessing of pulp-related physiological signals.

[0045] In some embodiments, the feature attention layer can process the set of higher-order feature representations through a self-attention mechanism and a fifth weight, thereby automatically selecting effective data in the set of higher-order feature representations that have a necessary connection with pulpitis; wherein, the fifth weight can be dynamically adjusted.

[0046] As can be seen from the above, in the clinical efficacy evaluation system for the vital pulp preservation treatment plan for permanent tooth pulpitis provided in this application embodiment, the preoperative evaluation module includes at least a dual-task regression neural network. The shared feature layer in the dual-task regression neural network is used to extract features from the preoperative physiological index set to obtain a high-order feature representation set of the preoperative physiological index set. In this way, the abstract semantic information in the preoperative physiological index set can be accurately extracted, so that the high-order feature representation set can accurately reflect the correlation and pattern between the data in the preoperative physiological index set. At the same time, through the feature attention layer in the dual-task regression neural network, redundant data in the high-order feature representation set is removed to obtain effective preoperative data for pulpitis. The effective preoperative data is associated with the preoperative state, thus reducing the redundancy of the effective preoperative data and improving the effectiveness and accuracy of the effective preoperative data.

[0047] Based on the foregoing embodiments, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in this application embodiment, the shared feature layer 104 includes a first hidden layer 106, a second hidden layer 107, and a third hidden layer 108; wherein: The first hidden layer 106 includes L neurons, which are used to map the set of preoperative physiological indicators to an L-dimensional feature vector; the second hidden layer 107 includes M neurons, which are used to extract and combine features from the L-dimensional feature vector to obtain an M-dimensional feature vector; the third hidden layer 108 includes N neurons, which are used to perform nonlinear processing and feature extraction on the M-dimensional feature vector to obtain a set of higher-order feature representations.

[0048] Among them, the L-dimensional feature vector is used to characterize the primary features of the preoperative physiological index set; the M-dimensional feature vector is used to characterize the intermediate features of the preoperative physiological index set; L, M, and N are all integers greater than 1, and the value of L is greater than M, and the value of M is greater than N.

[0049] In some embodiments, primary features can characterize the local features or local details of various indicators in the preoperative physiological indicator set, and the semantic information they carry may be less than that of intermediate features. Correspondingly, intermediate features can be obtained by associating and integrating the local features or local details in the primary features. Therefore, intermediate features can characterize the combination rules between the local features or local details represented by the preoperative physiological indicator set.

[0050] In some embodiments, L can be 64, so the first hidden layer 106 can contain 64 neurons. These neurons can receive four-dimensional data represented by the preoperative physiological index set and map the four-dimensional data to a 64-dimensional feature space to obtain 64-dimensional primary features. For example, after the first hidden layer 106 receives the preoperative physiological index set, a nonlinear unit such as ReLU can be used as the activation function to introduce nonlinearity and accelerate the convergence of the mapping operation. For example, in order to reduce the probability of overfitting, a Dropout layer can be connected after the first hidden layer 106 and its dropout rate can be set to 0.2. At the same time, L2 regularization is applied to the weight parameters of this layer, and its regularization coefficient can be set to 0.01 to penalize excessive weights and enhance the generalization ability of the first hidden layer.

[0051] In some embodiments, M can be 32, so the second hidden layer can contain 32 neurons. Through these neurons, the primary features output by the first hidden layer can be further combined to extract more complex 32-dimensional intermediate features. For example, at this time, the 32-dimensional intermediate features output by the second hidden layer can also be subjected to feature dimensionality reduction and abstraction processing by ReLU and L2 regularization, and its regularization coefficient can also be set to 0.01.

[0052] In some embodiments, N can be 16. In this case, the third hidden layer can contain 16 neurons. By performing nonlinear processing and feature extraction on the 32-dimensional intermediate features, a 16-dimensional shared feature vector can be obtained. This shared feature vector is the higher-order feature representation in the higher-order feature representation set. For example, in the third hidden layer, the Gaussian Error Linear Unit (GELU) can be used as the activation function. Compared with ReLU, the activation curve of GELU is smoother, and it has non-zero values ​​in the negative region, thus it can better capture the complex nonlinear dependencies between features in the 32-dimensional intermediate features. Therefore, this function is particularly suitable for processing medical data with physiological relevance.

[0053] As can be seen from the above, in the clinical efficacy evaluation system for the vital pulp preservation treatment plan for permanent tooth pulpitis provided in this application embodiment, the first hidden layer of the shared feature layer includes L neurons, which are used to map the set of preoperative physiological indicators to an L-dimensional feature vector. The L-dimensional feature vector is used to characterize the primary features of the set of preoperative physiological indicators, thus realizing the tracking and mapping of the primary features carried in the set of preoperative physiological indicators. Furthermore, through the second hidden layer of the shared feature layer, feature extraction and combination processing of the L-dimensional feature vector are performed to obtain an M-dimensional feature vector used to characterize the intermediate features of the set of preoperative physiological indicators, thus realizing targeted feature extraction and combination of primary features. On the other hand, by using the N neurons in the third hidden layer of the shared feature layer to perform nonlinear processing and feature extraction on the M-dimensional feature vector, a high-order feature set representation set is obtained. This enables the stepwise and progressive extraction of features carried in the preoperative physiological indicator set, thereby improving the comprehensiveness and completeness of the high-order feature set representation set. At the same time, since the value of L is greater than M, and the value of M is greater than N, a stepwise dimensionality reduction is also achieved from primary features to intermediate features and even high-order feature set representation sets. This reduces the amount of computation generated when processing the high-order feature set representation set in the subsequent process, thereby improving the efficiency of processing the high-order feature set representation set in the subsequent process.

[0054] Based on the foregoing embodiments, in the clinical efficacy evaluation system for the vital pulp preservation treatment plan for permanent tooth pulpitis provided in this application embodiment, the feature attention layer 105 is used to perform linear transformation on the high-order feature representation set through P N-dimensional linear transformation weight matrices to obtain the preoperative vector set; the feature attention layer 105 is also used to perform weighted representation on the vectors in the preoperative vector set through a self-attention mechanism to obtain the preoperative weighted result, and perform normalization processing on the preoperative weighted result to obtain the effective preoperative data.

[0055] The preoperative vector set includes query vectors, key vectors, and value vectors; P and N are both integers greater than 1.

[0056] In some embodiments, the P N-dimensional linear transformation weight matrices can be different, where P can be 3, and the three N-dimensional linear transformation weight matrices can be continuously optimized and adjusted through model training; for example, the three N-dimensional linear transformation weight matrices can be denoted as Wq, Wk and Wv.

[0057] In some embodiments, the query vector (Q) may include key features required by the feature attention layer in the current task; the key vector (K) may include feature labels for comparison and matching with the query vector, and the feature labels are used to characterize the content represented by the key features; the value vector (V) may characterize the actual value of each feature obtained from the set of higher-order feature representations; it should be noted that the query vector, key vector and value vector may have the same dimension.

[0058] In some embodiments, the preoperative vector set can be obtained in the following way: Obtain the set of higher-order feature representations H from the output of the shared feature layer. Then, input the higher-order feature representations in the 16-dimensional set of higher-order feature representations into three independent linear transformation weight matrices, and generate three query vectors, key vectors, and value vectors, all of which are 16-dimensional, through matrix multiplication.

[0059] In some embodiments, the attention weight matrix upon which the weighted representation depends can be obtained in the following way: First, consider the transpose of the query vector P and the key vector K. Perform matrix multiplication to obtain a 16×16 first matrix, where each element reflects the degree of matching between each query in P and each key in K; then, divide each element of the first matrix by the square root of the dimension of the key vector. The second matrix is ​​obtained; this step is used to alleviate the problem of excessively large values ​​caused by high vector dimensions, in order to stabilize gradient propagation. Let K be the dimension of the key vector. Since the dimension of the shared feature vector is 16, ... The value of is 16; then, the second matrix is ​​normalized using the Softmax function to generate a 16×16 attention weight matrix. Specifically, the above process can be implemented using equation (2): (2) It should be noted that each row in the attention weight matrix can correspond to one of the higher-order feature representations in the set of higher-order feature representations H, and the 16 values ​​in each row represent the importance score of that feature relative to the other 15 features and itself.

[0060] Accordingly, the preoperative weighted result can be obtained by performing matrix multiplication between the attention weight matrix and the value vector.

[0061] It should be noted that, in the embodiments of this application, the self-attention mechanism can be implemented through an attention weight matrix. In this way, the attention weight matrix enables the feature attention layer to learn and track the correlation between different features in the set of higher-order feature representations, thereby realizing the differentiated dynamic processing of different features in the set of higher-order feature representations. This enhances the correlation between the preoperative weighted results and the subsequent data processing process, and can also effectively suppress noise and redundant data in the preoperative vector set.

[0062] Specifically, when the feature attention layer processes high-order feature representations that are strongly correlated with inflammation, such as high MPI coefficient of variation and low TOI, it can dynamically assign higher weights to them through the attention weight matrix, while reducing the weights of irrelevant noise features, including small measurement error signals. This allows key information to be obtained from the aforementioned high-order feature representations. Furthermore, it can suppress redundant interference and provide more accurate and effective data input for the subsequent dual-task dedicated layer.

[0063] As can be seen from the above, in the clinical efficacy evaluation system for the vital pulp preservation treatment scheme for permanent tooth pulpitis provided in this application embodiment, the feature attention layer is used to perform linear transformations on the higher-order feature representations in the higher-order feature representation set through P N-dimensional linear transformation weight matrices, to obtain a preoperative vector set including query vectors, key vectors, and value vectors. In this way, the effective information in the higher-order feature representation set can be preserved. On this basis, the feature attention layer is also used to perform weighted representations on the vectors in the preoperative vector set based on a self-attention mechanism to obtain preoperative weighted results. In this way, targeted screening of key data in the preoperative vector set can be achieved, and the redundancy of the preoperative weighted results can be reduced. On the other hand, by normalizing the preoperative weighted results, the range of values ​​of the preoperative effective data can be narrowed, thereby reducing the computational and storage resource requirements for subsequent processing of preoperative effective data.

[0064] Based on the foregoing embodiments, the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in this application embodiment further includes a dual-task regression neural network with a dual-task dedicated layer 109; the dual-task dedicated layer 109 includes an inflammation processing unit 110 and a vitality processing unit 111 arranged in parallel; wherein: The inflammation processing unit 110 is used to perform nonlinear mapping on the preoperative effective data associated with the set of higher-order feature representations to obtain the inflammation status; the vitality processing unit 111 is used to extract data associated with the tissue function of the dental pulp from the preoperative effective data to obtain the vitality status.

[0065] In some embodiments, the inflammation processing unit 110 can extract inflammation-related data related to the dental pulp from the preoperative effective data and quantify the inflammation-related data. For example, the inflammation processing unit 110 can perform nonlinear processing on the preoperative effective data using ReLU to obtain a first processing result, and then map the first processing result through a first fully connected layer including 8 neurons to accurately extract the core inflammatory features most related to inflammation from the first processing result. Then, the core inflammatory features are processed into a first probability value through Sigmoid, and the first probability value is linearly scaled to an I-score with a value range of [0, 100] to intuitively reflect the severity of inflammation. The inflammatory state can be represented by the I-score, and the core inflammatory features can include metabolic stress fluctuations and blood flow disturbance signals, etc.

[0066] In some embodiments, the structure of the vitality treatment unit may be symmetrical to that of the inflammation treatment unit, but its weight parameters are trained independently relative to the inflammation treatment unit. This unit focuses on assessing pulp viability.

[0067] Specifically, through the ReLU in the vitality processing unit and the second fully connected layer composed of 8 neurons, key vitality features related to cell survival and tissue function can be extracted from the effective preoperative data. Then, the key vitality features are output as a second probability value through the Sigmoid function. The second probability value is then linearly mapped to a V-score with a value range of [0, 100] to quantify the physiological function and survival potential of the dental pulp, achieving parallel and synergistic output with the I-score. Among them, the key vitality features may include cell membrane integrity, oxygenation stability, etc.

[0068] Figure 3 This is a schematic diagram of the structure of the dual-task regression neural network provided in the embodiments of this application, as shown below. Figure 3As shown, the dual-task regression neural network 103 may include an input layer, a shared feature layer, a feature attention layer, a dual-task dedicated layer, and an output layer arranged in cascade. The input layer is used to input and preprocess dental pulp-related physiological signals to obtain preoperative physiological indicators. The output layer can output the I-score and V-score from the dual-task dedicated layer. Furthermore, the inflammation head in the dual-task dedicated layer can correspond to the inflammation processing unit in the aforementioned embodiment, and the vitality head can correspond to the vitality processing unit in the aforementioned embodiment.

[0069] For example, the dual-task regression neural network can be pre-trained based on preoperative samples; wherein, the preoperative samples may include a set of preoperative physiological indicators of pulpitis from multiple clinical cases; for example, before training begins, a loss function as shown in equation (3) can be constructed: (3) in, Indicates the inflammation prediction score, This represents the true score of inflammation. Indicates the vitality prediction score, The first two terms, I-score and V-score, are calculated using mean squared error (MSE) to determine the prediction accuracy. The third term is a coupling constraint term to ensure that the predicted data show a biological negative correlation between low vitality and high inflammation.

[0070] Then, for the training process of the dual-task regression neural network, a training strategy including a first stage and a second stage can be formulated. For the first stage, a pre-training based on 1200 pathological data and 50 rounds of optimization can be used, freezing the shared feature layer and adjusting only the parameters of the dual-task-specific layer. For the second stage, the shared feature layer can be unfrozen, and the parameters of the dual-task-specific layer can be fine-tuned for 100 rounds using the full set of preoperative samples. The above process can be implemented using the AdamW optimizer with a learning rate of 0.001 and a weight decay of 0.0001, and the optimal parameters can be saved through an early stopping mechanism.

[0071] It should be noted that preoperative samples must be obtained with the patient's permission.

[0072] The above structure enables integrated configuration of data input, feature extraction, preoperative state determination, and output of the dual-task regression neural network, achieving integrated and targeted processing of dental pulp-related physiological signals.

[0073] As can be seen from the above, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment scheme provided in this application embodiment, the dual-task regression neural network further includes a dual-task dedicated layer. This dual-task dedicated layer includes a parallel inflammation processing unit and a vitality processing unit. The inflammation processing unit is used to perform nonlinear mapping on the preoperative effective data associated with the high-order feature representation set to obtain the inflammation state. The vitality processing unit is used to extract data related to the tissue function of the dental pulp from the preoperative effective data to obtain the vitality state. Thus, parallel extraction of the inflammation state and vitality state is achieved, improving the processing efficiency of preoperative effective data. Moreover, by setting the inflammation processing unit and the vitality processing unit in parallel, the inflammation state and vitality state can be extracted separately, thereby reducing the coupling between the two extraction processes and improving the accuracy of the inflammation state and vitality state.

[0074] Based on the foregoing embodiments, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in this application embodiment, the preoperative evaluation module 101 includes a feature map construction unit 112, which is used to construct and display a two-dimensional preoperative feature map based on the inflammation status and vitality status.

[0075] The two-dimensional preoperative feature map includes at least a low-risk area, a reversible inflammation area, a borderline area, and an irreversible inflammation area.

[0076] In some embodiments, the two-dimensional preoperative feature map can be constructed in a two-dimensional coordinate system with the I-score corresponding to the inflammation state as the horizontal axis and the V-score corresponding to the vitality state as the vertical axis, wherein the values ​​of I-score and V-score can both be in the range of [0, 100].

[0077] In some embodiments, the low-risk area, reversible inflammation area, critical area, and irreversible inflammation area can be divided in the above two-dimensional coordinate system according to a preset threshold, and can also be distinguished by different colors.

[0078] Specifically, low-risk areas can be displayed in green, and can specifically correspond to the range where I-score is less than 20 and V-score is greater than 80. For example, low-risk areas can indicate that the pulp is in a healthy or low-risk state. For instance, a low-risk area can indicate that the pulp blood perfusion is normal, oxygenation is sufficient, cell membranes are intact, metabolism is stable, and there is no obvious inflammatory response. At this time, the pulp is in a healthy or near-healthy state, and there is no need to perform a pulp preservation treatment plan. Only superficial caries and other causes need to be repaired and regular follow-up is set.

[0079] Specifically, reversible inflammation areas can be represented by yellow, which can correspond to an I-score greater than or equal to 20 and less than 50, while a V-score greater than 60. For example, a reversible inflammation area can indicate that there is a mild to moderate inflammatory reaction in the gingiva, but the pulp tissue is viable and the cell membrane has not suffered irreversible damage. The inflammation is reversible, and a vital pulp preservation treatment can be implemented at this time. Among them, moderate inflammatory reactions can include, for example, pulp hyperemia.

[0080] Specifically, the critical region can also be called the observation region, which can be displayed in gray; for example, the setting conditions for the critical region can include any of the following: (1) Condition 1: I-score is greater than or equal to 50 and less than 70, and V-score is greater than or equal to 40 and less than or equal to 60; (2) Condition 2: I-score is greater than or equal to 20 and less than 50, and V-score is greater than or equal to 40 and less than or equal to 60.

[0081] The critical zone can indicate a moderate degree of inflammation or a critical level of pulp vitality. It is not possible to immediately determine whether the inflammation is reversible. Further observation is required in conjunction with clinical symptoms. If the inflammation subsequently shifts from the critical zone to a reversible inflammation zone, the corresponding procedure is performed. If the inflammation subsequently shifts from the critical zone to an irreversible inflammation zone, root canal treatment is performed.

[0082] Specifically, the irreversible inflammatory area can also be called the necrotic area, which can be displayed in red. For example, the irreversible inflammatory area can be configured with the following conditions: I-score greater than or equal to 70 or V-score less than 40. The irreversible inflammatory area can indicate a severe inflammatory response or severe loss of pulp vitality, extensive cell membrane damage, irreversible inflammation, or even pulp necrosis. In this case, pulp preservation treatment can be abandoned, and root canal treatment can be performed directly. Severe inflammatory response can include acute purulent pulpitis.

[0083] It should be noted that the thresholds used to determine the range of I-score and V-score values ​​can be fixed or adjusted, and this application embodiment does not limit this.

[0084] As described above, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment provided in this application embodiment, the preoperative evaluation module includes a feature map construction unit, which is used to construct and display a two-dimensional preoperative feature map based on the inflammation state and vitality state. The two-dimensional preoperative feature map includes at least a low-risk area, a reversible inflammation area, a borderline area, and an irreversible inflammation area. Thus, through the feature map construction unit, a comprehensive and intuitive display of the inflammation state and vitality state can be achieved.

[0085] Based on the foregoing embodiments, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in this application embodiment, the postoperative efficacy evaluation module 102 includes a weighted Bayesian network 113, which is used to evaluate the importance of indicators in the postoperative physiological indicator set based on postoperative weights, and to process the postoperative physiological indicators based on a Conditional Probability Table (CPT) to obtain CES.

[0086] In some embodiments, postoperative weights can be pre-constructed based on postoperative samples to characterize the degree of influence of physiological indicators in the postoperative physiological indicator set on CES; for example, postoperative samples and preoperative samples can be correlated with each other.

[0087] In some embodiments, postoperative samples can be obtained in the following ways: First, 1000 cases of patients undergoing vital pulp preservation treatment for permanent tooth pulpitis were obtained. These 1000 cases were then binary-labeled to obtain labeled samples. The labeled samples could include a first set of samples with successful clinical outcomes, where postoperative pulp pathology showed viable pulp, CBCT showed an intact dentin bridge, and a SII greater than or equal to 90. A second set of samples with unsuccessful clinical outcomes could also be included, where postoperative pulp pathology showed non-viable pulp, CBCT showed an incomplete dentin bridge, and an SII less than 90. These patient cases could be obtained from the dental departments of at least three tertiary hospitals, with the permission of both the patients and the hospitals.

[0088] Secondly, the labeled samples are associated with the corresponding FSI, SRI, and SII values ​​to construct a structured dataset. For example, the labeled samples can be divided into a training set, a validation set, and a test set in a 7:2:1 ratio. The training set is used for parameter tuning, the validation set is used for parameter optimization, and the test set is used for final performance evaluation.

[0089] Next, feature discretization is performed on the labeled samples. Specifically, the FSI, SRI, and SII values ​​in the training, validation, and test sets can be uniformly discretized into values ​​ranging from [0, 10] at 10-point intervals to eliminate dimensional differences and thus obtain postoperative samples; for example, the above discretization process may include discretizing 91-100 into 10. This step aims to ensure consistent input format.

[0090] For example, postoperative weights can be constructed as follows: First, for all cases included in the postoperative sample, the clinical outcome (success or failure) is taken as the target variable Y, and the FSI, SRI, and SII in the postoperative sample are taken as the feature variables X. The information entropy of the target variable is calculated using equation (4). To quantify the overall uncertainty of clinical outcomes: (4) in, The percentage of cases with a successful clinical outcome. This represents the proportion of cases where the clinical outcome was unsuccessful.

[0091] Secondly, calculate the residual uncertainty of the target variable Y after knowing the value of a certain feature, such as FSI. Specifically, taking FSI as an example, for each of its 10 discretized intervals vv, calculate the conditional entropy H(Y|FSI=vv) of the samples in that interval. Then, according to the proportion p(vv) of the number of samples in each interval vv to the postoperative samples, sum the conditional entropies of all intervals by weight to obtain the overall conditional entropy of FSI as shown in equation (5): (5) Correspondingly, the conditional entropies of SRI and SII are calculated in the same way, and the information gain is then calculated using equation (6) to intuitively express the reduction in uncertainty of the target variable Y after obtaining information about the feature Xᵢ corresponding to the i-th postoperative physiological index: (6) Where i is an integer greater than or equal to 1 and less than or equal to 3; The larger the value, the greater the contribution of the feature to the clinical outcome of treatment success or failure, that is, the higher the importance of the feature; finally, the information gain of the three features is normalized to obtain the weight value of each feature. For example, the weight of FSI can be calculated by equation (7): (7) In some embodiments, the magnitude of the postoperative weight of a postoperative physiological indicator in the set of postoperative physiological indicators can be used to assess the importance of the corresponding postoperative physiological indicator to CES.

[0092] In some embodiments, a weighted Bayesian network may include a cascaded feature layer and a target layer.

[0093] For example, the feature layer, also known as the input layer, may include three independent discrete nodes, which are used to correspond to the inputs FSI, SRI, and SII, respectively, so that they are discrete into postoperative discrete indicators at intervals of 10. The value range of the postoperative discrete indicators is [1, 10]. In this way, the dimensional differences between indicators in the postoperative physiological indicator set can be eliminated.

[0094] For example, the target layer can have only one continuous output node to receive postoperative discrete indicators and process them in conjunction with CPT to obtain the quantized probability value corresponding to the postoperative discrete indicators. Then, the quantized probability value is linearly processed to obtain CES. For example, the quantized probability value can characterize the posterior probability of treatment success.

[0095] In some embodiments, only a one-way direct connection from the feature layer to the target layer is maintained between the feature layer and the target layer, without any cross connections or hierarchical jump connections between features. In this way, FSI, SRI, and SII can each independently point to the treatment success probability node, and cross connections can include the connection between FSI and SRI. Through the above-mentioned connection relationship between the feature layer and the target layer, the logic of parallel assessment of function, structure, and symptoms in clinical practice can be aligned, reducing fictitious causal relationships between different indicators, reducing the model parameter size, lowering the difficulty of constructing CPT, and intuitively tracing the correlation path between the set of postoperative physiological indicators and the probability results, thereby improving the clinical interpretability of the reasoning process.

[0096] In some embodiments, the weighted Bayesian network can be pre-trained based on postoperative samples; specifically, the training of the weighted Bayesian network can be performed synchronously with the iterative optimization of CPT.

[0097] For example, the initial CPT can be iteratively optimized and refined based on the validation set of postoperative samples. During this process, the optimization objective is to minimize the model's MSE on the validation set, and unreliable probability entries corresponding to feature combinations with insufficient sample size in the training set are smoothed using a neighboring combination probability mean filling strategy. This optimization process can be iterated repeatedly until the MSE converges to below the target threshold on the validation set, at which point a fully validated and refined weighted Bayesian network can be obtained.

[0098] For example, a feature combination with too few samples may include a combination of postoperative physiological indicators with a sample size of less than or equal to 5, and the target threshold may be 0.05.

[0099] For example, after obtaining the weighted Bayesian network through the above training process, its prediction accuracy can be tested based on the test set. Only when its accuracy is greater than or equal to 85% and the area under the receiver operating characteristic curve (ROC) (AUC) is greater than or equal to 0.88 can it be confirmed that the robustness and reliability of the Bayesian network model meet the actual data processing needs.

[0100] In some embodiments, the CPT can be pre-constructed based on postoperative samples, and the CPT can characterize all possible combinations of postoperative physiological indicators in the postoperative samples, as well as the true probability of clinical success for each combination; for example, a three-dimensional CPT can be constructed based on postoperative samples and maximum likelihood estimation.

[0101] Specifically, we can iterate through the feature combinations corresponding to the 1000 sets of postoperative physiological indicators in the postoperative sample (10×10×10), and count the total number of cases corresponding to each combination in the postoperative sample, as well as the number of cases with successful clinical outcomes in the postoperative sample. The conditional probability of a successful clinical outcome is calculated by the quotient between the number of successful cases and the total number of cases. For example, we can count the 700 training cases included in the postoperative sample one by one to obtain 20 cases with all FSI values ​​of 9, SRI values ​​of 9, and SII values ​​of 10. If 19 of these 20 cases have successful clinical outcomes, then the success probability corresponding to this combination is the quotient of 19 and 20, which is 0.95. By counting the 1000 combinations included in the postoperative sample in the above way, we can obtain the CPT represented by a three-dimensional table.

[0102] In some embodiments, CES is obtained by processing the set of physiological indicators based on CPT, which can be achieved in the following ways: If there is a feature combination in CPT that perfectly matches the set of postoperative physiological indicators, then the probability corresponding to the set of postoperative physiological indicators in CPT can be determined as a quantitative probability value, and CES can be determined based on the quantitative probability value using the above method.

[0103] In some embodiments, CES can be obtained by linearly processing the quantified probability value to a range of [0, 100]. For example, the quantified probability value can be multiplied by 100 to obtain CES. For example, after obtaining CES, the CES can be divided into intervals and corresponding clinical recommendations can be output. For example, if CES is greater than or equal to 85, the clinical effect of the viable pulp preservation treatment is excellent. In this case, the clinical recommendation may include a recommendation for long-term follow-up. If the value of CES is greater than or equal to 70 but less than or equal to 84, the clinical effect can be good. In this case, the clinical recommendation may include a recommendation to adjust the follow-up period or treatment intervention. If the value of CES is greater than or equal to 50 but less than or equal to 69, the clinical effect can be a warning, i.e., there is a risk of failure. In this case, the clinical recommendation may include: recommending to shorten the follow-up period to 1 to 3 months and to specifically review abnormal indicators. If the value of CES is less than or equal to 49, the clinical effect can be confirmed as failure. In this case, the clinical recommendation may include: terminating the current treatment plan and switching to root canal treatment.

[0104] As can be seen from the above, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment scheme provided in this application embodiment, the postoperative efficacy evaluation module includes a Bayesian weighted network, which is used to evaluate the importance of indicators in the postoperative physiological indicator set based on postoperative weights, thereby quantifying the degree of influence of the postoperative physiological indicator set on the final CES; and by processing the postoperative physiological indicator set based on CPT to obtain the CES, the accurate mapping of the postoperative physiological indicator set can be achieved, thereby simplifying the CES determination process and improving the accuracy of CES.

[0105] Based on the foregoing embodiments, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in this application embodiment, the postoperative efficacy evaluation module is also used to correct the CES based on postoperative weights.

[0106] In some embodiments, CES can be modified in the following ways: If the weight corresponding to a certain postoperative physiological indicator in the postoperative weight is greater than the weight threshold, but the value of the postoperative physiological indicator is less than 60, the CES can be corrected by lowering it; if the value of the postoperative physiological indicator corresponding to the above weight is greater than or equal to 90, the CES can be corrected by raising it, so as to increase the explicitness of the influence of the key postoperative physiological indicators corresponding to high weight.

[0107] For example, the magnitude and range of CES adjustment can be determined by dental professionals.

[0108] As can be seen from the above, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in this application embodiment, the postoperative efficacy evaluation module is also used to correct the CES based on the postoperative weight. In this way, the CES can be dynamically and flexibly corrected, thereby improving the clinical accuracy of the CES.

[0109] Based on the foregoing embodiments, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in this application embodiment, the postoperative efficacy evaluation module further includes a weighted Bayesian network, which is used to process the set of postoperative physiological indicators through K-nearest neighbor interpolation to obtain CES.

[0110] In some embodiments, if there is no index combination in CPT that perfectly matches the postoperative physiological index set, the postoperative physiological index set can be processed by K-nearest neighbor interpolation to obtain CES.

[0111] In some embodiments, K can be 3, and the average probability of neighboring combinations can be taken as the quantized probability value. The quantized probability value is then linearly processed using the method provided in the foregoing embodiments to obtain CES.

[0112] As can be seen from the above, in the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment scheme provided in this application embodiment, the postoperative efficacy evaluation module includes a weighted Bayesian network, which is used to process the set of postoperative physiological indicators through K-nearest neighbor interpolation to obtain CES. In this way, a diversified process for determining CES is provided, thereby improving the flexibility and robustness of the postoperative efficacy evaluation module in determining CES.

[0113] Based on the foregoing embodiments, the clinical efficacy evaluation system for the permanent tooth pulpitis vital pulp preservation treatment plan provided in this application embodiment further includes a parameter update module 114, which is used to update the postoperative weight of the postoperative efficacy evaluation module based on the updated postoperative sample if a postoperative sample update is detected.

[0114] Accordingly, if no postoperative sample update is detected, the postoperative weights do not need to be updated.

[0115] For example, to improve the consistency between postoperative weights and long-term clinical practice and to improve the accuracy of postoperative weights, a regular iterative update mechanism for postoperative weights can be established in advance. Specifically, when 500 new postoperative follow-up cases have been collected, and the postoperative follow-up cases include data on FSI, SRI, SII and one-year clinical outcomes that are successful, it can be determined that the postoperative sample has been updated, and the iterative process can be started at this time.

[0116] For example, the iterative process may include all the steps for determining postoperative weights in the foregoing embodiments.

[0117] For example, the method provided in the foregoing embodiments can also be used to update the CPT based on the updated postoperative sample.

[0118] As described above, in the clinical efficacy evaluation system for the vital pulp preservation treatment plan for permanent tooth pulpitis provided in this application embodiment, the parameter update module of the clinical efficacy evaluation system is used to update the postoperative weights of the postoperative efficacy evaluation module based on the updated postoperative samples if a postoperative sample update is detected. This achieves continuous iterative updates of the postoperative weights, thereby improving the accuracy of the postoperative weights and enabling them to meet diverse case characteristics and changes in clinical scenarios, thus improving the accuracy of the postoperative efficacy evaluation module.

[0119] The following are specific application examples of the clinical efficacy evaluation system provided in the embodiments of this application: The patient was a 25-year-old female who presented with pain in her upper right posterior teeth for 3 days due to hot and cold stimuli. She had no spontaneous pain. Clinical examination revealed deep caries on the occlusal surface of the upper right first molar, which was sensitive to probing and tender to percussion. There was no redness or swelling of the gums.

[0120] Preoperative diagnosis included: reversible pulpitis of permanent teeth, with a proposed vital pulp preservation treatment, and the clinical efficacy evaluation system provided in this application was used to assist in decision-making throughout the process.

[0121] First, the preoperative assessment process can involve the following steps: First, the acquisition and standardization of multimodal dental pulp-related physiological signals. Specifically, this may include: Physiological signals related to the dental pulp were acquired using acquisition equipment. This included: continuous acquisition of pulp blood perfusion signals via LDF and conversion to PU values, calculating the pulp blood perfusion volume (average PU value 65) and standard deviation (8.5); simultaneous acquisition of blood flow pulsation waveforms, measuring the peak value to 80 and the trough value to 50; acquisition of red and infrared light intensity changes following the pulse after passing through tissue using a pulse oximeter, calculating arterial oxygen saturation (SO2 98%) using a dedicated algorithm; acquisition of raw impedance data containing multiple frequency points, including impedance modulus (average 500Ω) and phase angle (average -15°), through EIS scanning of the dental pulp at frequencies from 10Hz to 1MHz; and acquisition of spectral intensity in the 700 to 1800nm ​​band using a near-infrared spectrometer, analyzing and retrieving HbO2 and Hb concentrations, and further calculating the Hb to HbT ratio (0.82).

[0122] Data preprocessing was performed: Kalman filtering was used to remove artifacts caused by gingival movement and retain the true blood flow pulsation characteristics for both LDF and blood oxygen data; adaptive iterative reweighted penalized least squares algorithm was used to remove baseline drift caused by enamel scattering for spectral data; and high-frequency noise was eliminated for impedance data by smoothing with a 5th-order polynomial.

[0123] Data completion and calibration were performed: Two frequency points in the EIS data were found to be anomalous. These were corrected using a Cole-Cole model, and the cell membrane capacitance (Cm = 1.2 μF / cm) was extracted from them. 2 Core physiological parameters such as extracellular fluid resistance (Re is 200Ω) were measured. In this case, there were no missing data accounting for less than 5%, so KNN interpolation was not required to complete the data.

[0124] Normalization: All preprocessed data are mapped to the [0, 1] interval using the Min-Max method.

[0125] Second, calculate preoperative physiological parameters. Specifically: For MPI: The pulsatility index was determined to be approximately (80-50) / 65≈0.46; The MPI value was determined to be (0.6×0.65+0.4×0.46)×100≈58 points (indicating moderate blood perfusion and average pulsatility).

[0126] Regarding TOI: The TOI score is determined to be approximately (0.7 × 0.82 + 0.3 × 0.98) × 100 ≈ 87 points (indicating sufficient oxygen supply).

[0127] For CMII: The value of CMII can be determined to be 0.92 × 100 = 92 points (Cm normalized value = 0.92, indicating that the cell membrane is intact).

[0128] For MSI: It can be determined that the MPI coefficient of variation is 0.15 (low), the TOI value is 0.87 (high), and Re=0.5 (medium). The fuzzy logic reasoning is "low stress", and the MSI value after defuzzification is 28.

[0129] Third, the preoperative physiological indicators were processed using a dual-task regression neural network. The input data for the dual-task regression neural network is a 4-dimensional vector [58, 87, 92, 28]. The first hidden layer, containing 64 neurons, processes this 4-dimensional vector through a shared feature layer, yielding 64 primary features. Dropout, associated with the first hidden layer, removes 12 redundant features. The second hidden layer, with 32 neurons, refines these into 32 intermediate features, which are then further refined by a third hidden layer to obtain 16 shared features. Next, a feature attention layer weights the features; for example, a weight of 0.3 is assigned to MSI-related features corresponding to the inflammation state, and a weight of 0.7 is assigned to CMII-related features associated with the vitality state. Then, the dual-task dedicated layer outputs the results. For instance, features refined using ReLU from the 8 neurons of the inflammation head are then output with a Sigmoid function of 0.32, resulting in an I-score of 32. Similarly, features refined using the 8 neurons of the vitality head are output with a Sigmoid function of 0.85, resulting in a V-score of 85.

[0130] Fourth, clinical decision-making. The I-score and V-score can be processed by the feature map construction unit to construct and display a two-dimensional preoperative feature map. For example, the two-dimensional preoperative feature map can determine that the current state of the patient is a reversible inflammatory area. Therefore, the viable bone marrow preservation treatment plan of indirect medullary capping combined with resin filling is recommended.

[0131] Fifth, implement a viable bone marrow preservation treatment plan that combines indirect medullary capping with resin filling.

[0132] Secondly, regarding the postoperative effect evaluation, the following steps can be performed: First, after 12 months, multi-source follow-up data were collected; specifically, functional data, CBCT images, and symptom data were included. For functional data, MPI-s (62) could be obtained through LDF acquisition, and TOI-s (85) could be obtained through near-infrared spectroscopy and pulse oximetry. For CBCT images, Digital Imaging and Communications in Medicine (DICOM) data could be obtained through CBCT scanning, and dentin bridges could be extracted based on Otsu threshold segmentation. For symptom data, the following data could be included by the attending physician: VAS pain score (0, preoperative value was 6), percussion pain grade was 0, occlusal discomfort was 1 and there was no subjective discomfort.

[0133] Second, a multi-dimensional set of postoperative physiological indicators was constructed.

[0134] Specifically: For FSI, the weighted sum of 62 and 85 with weights (0.6, 0.4) yields 71.2, and the nearest value is 71. For SRI, the ratio of dentin bridge thickness to pulp chamber floor diameter is 0.8, and the ratio of dentin bridge density to normal dentin density is 0.95. The weighted sum of 0.8 and 0.95 with weights (0.5, 0.5) yields 0.875. Meanwhile, Canny edge detection shows that the filling gap is 20μm. Therefore, the value of SRI is the product of 0.875 and 100, that is, the value of SRI is approximately 88. For SII, its value of 100 can be calculated by (1-0 / 6)×0.6+(1-0 / 3)×0.3+1×0.1.

[0135] Third, a multi-indicator fusion comprehensive evaluation is performed on the postoperative physiological index set using a weighted Bayesian network. The postoperative weights can include (0.4, 0.3, 0.3), and the values ​​of FSI, SRI, and SII are 7, 9, and 10, respectively. At this time, based on the corresponding combination in CPT, the quantitative probability value is determined to be 0.92, thus determining the value of CES to be 92, which means the clinical effect is excellent. The corresponding clinical recommendations can include: switch to annual routine follow-up without special intervention.

[0136] Furthermore, the postoperative weights can be iteratively updated via the parameter update module.

[0137] Specifically, the conditions for triggering iterative updates may include: collecting a cumulative total of 600 new cases within 2 years post-surgery; the iterative update process may include: recalculating the information gain to adjust the weight corresponding to FSI from 0.4 to 0.42, the weight corresponding to SRI from 0.35 to 0.33, and the weight corresponding to SII from 0.25 to 0.23; and reconstructing CPT to add combinations of postoperative physiological indicators not included in the CPU. After the above iterative updates, the accuracy of the postoperative effect assessment module can be improved to 89%.

[0138] The clinical efficacy evaluation system provided in this application embodiment can achieve an effective and precise closed loop for preoperative assessment of pulpitis status, implementation of vital pulp preservation treatment plan, and postoperative clinical efficacy evaluation for any patient.

[0139] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0140] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0141] In the several embodiments provided in this application, it should be understood that the disclosed device can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0142] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0143] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0144] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new system embodiments.

[0145] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A system for evaluating the clinical effectiveness of a treatment protocol for the vital-pulp-sparing treatment of permanent dental pulpitis, characterized in that it comprises: The method comprises the following steps: a preoperative evaluation module is used to obtain a set of preoperative physiological indicators and process the set of preoperative physiological indicators to obtain a preoperative state of pulpitis of the permanent tooth, so as to determine whether to implement the vital-pulp-salvage treatment scheme for the permanent tooth; wherein the set of preoperative physiological indicators comprises a microcirculation perfusion index MPI, a tissue oxygenation index TOI, a cell membrane integrity index CMII and a metabolic stress index MSI; the preoperative state comprises an inflammation state and a vitality state of the permanent tooth; a postoperative effect evaluation module is used to obtain a set of postoperative physiological indicators and comprehensively evaluate the set of postoperative physiological indicators to obtain a clinical comprehensive treatment effect score CES of the pulpitis within at least one period after the vital-pulp-salvage treatment scheme ends; wherein the set of postoperative physiological indicators comprises a pulp function survival index FSI, a structure repair quality index SRI and a clinical symptom improvement index SII.

2. The clinical effect evaluation system according to claim 1, characterized by, The preoperative evaluation module comprises at least a double-task regression neural network; the double-task regression neural network comprises at least a shared feature layer and a feature attention layer; wherein: the shared feature layer is used to extract features of the set of preoperative physiological indicators to obtain a set of high-order feature representations of the set of preoperative physiological indicators; the feature attention layer is used to remove redundant data in the set of high-order feature representations to obtain preoperative effective data of the pulpitis; wherein the preoperative effective data is associated with the preoperative state.

3. The clinical effect assessment system according to claim 2, characterized in that, The shared feature layer comprises a first hidden layer, a second hidden layer and a third hidden layer; wherein: the first hidden layer comprises L neurons and is used to map the set of preoperative physiological indicators to an L-dimensional feature vector; wherein the L-dimensional feature vector is used to represent primary features of the set of preoperative physiological indicators; the second hidden layer comprises M neurons and is used to extract and combine features of the L-dimensional feature vector to obtain an M-dimensional feature vector; wherein the M-dimensional feature vector is used to represent intermediate features of the set of preoperative physiological indicators; the third hidden layer comprises N neurons and is used to perform nonlinear processing and feature extraction processing on the M-dimensional feature vector to obtain the set of high-order feature representations; wherein L, M and N are all integers greater than 1, the value of L is greater than M, and the value of M is greater than N.

4. The clinical effect assessment system according to claim 2, characterized in that, The feature attention layer is used to perform linear transformation on high-order feature representations in the set of high-order feature representations through P N-dimensional linear transformation weight matrices to obtain a set of preoperative vectors; wherein the set of preoperative vectors comprises a query vector, a key vector and a value vector; P and N are both integers greater than 1; The feature attention layer is also used to perform weighted representation on vectors in the set of preoperative vectors through a self-attention mechanism to obtain a preoperative weighted result, and perform normalization processing on the preoperative weighted result to obtain the preoperative effective data.

5. The clinical effect assessment system of claim 2, wherein, The double-task regression neural network further comprises a double-task exclusive layer; the double-task exclusive layer comprises an inflammation processing unit and a vitality processing unit arranged in parallel; wherein: The inflammation processing unit is configured to perform nonlinear mapping on preoperative valid data associated with the high-order feature representation set to obtain the inflammation state. The vitality processing unit is configured to extract data associated with tissue function of the dental pulp from the preoperative valid data to obtain the vitality state.

6. The clinical effect assessment system of claim 1, wherein, The preoperative assessment module includes a feature map construction unit configured to construct and display a two-dimensional preoperative feature map based on the inflammation state and the vitality state. The two-dimensional preoperative feature map at least includes a low-risk area, a reversible inflammation area, a critical area, and an irreversible inflammation area.

7. The clinical effect assessment system of claim 1, wherein, The postoperative effect assessment module includes a weighted Bayesian network configured to evaluate the importance of postoperative physiological indicators in the postoperative physiological indicator set based on postoperative weights, and process the postoperative physiological indicator set based on a conditional probability table to obtain the CES.

8. The clinical effect evaluation system according to claim 7, characterized by, The postoperative effect assessment module is further configured to correct the CES based on the postoperative weights.

9. The clinical effect assessment system of claim 1, wherein, The postoperative effect assessment module includes a weighted Bayesian network configured to process the postoperative physiological indicator set by a K-nearest neighbor interpolation method to obtain the CES.

10. The clinical effect evaluation system according to any one of claims 1 to 9, characterized by, The clinical effect assessment system further includes a parameter updating module configured to, if it is detected that a postoperative sample is updated, update at least the postoperative weights of the postoperative effect assessment module based on the updated postoperative sample.

Images