Automatic generation system and method of tooth extraction diagnosis scheme

Abstract

The invention discloses an automatic generation system and method of a tooth extraction diagnosis scheme, and relates to the technical field of tooth extraction diagnosis, patient data is collected through a CBCT image data acquisition module, standardization, denoising and contrast enhancement processing are carried out through an image preprocessing module, and the image quality is improved. And then, a tooth image segmentation module is used for implementing fine segmentation based on boundary enhancement on the preprocessed image so as to accurately separate a tooth structure. The tooth feature quantitative extraction module further extracts multi-dimensional feature parameters from the segmentation result, wherein the multi-dimensional feature parameters comprise key indexes such as tooth root length and curvature. And finally, the intelligent tooth extraction diagnosis module automatically gives out a tooth extraction difficulty assessment and main risk prediction conclusion according to the extracted characteristic parameters in combination with a clinical knowledge rule base, and a structured diagnosis report generation module integrates all related information to form comprehensive diagnosis suggestions, so that the accuracy and efficiency of tooth extraction diagnosis are remarkably improved.

Description

Technical Field

[0001] This application relates to the technical field of tooth extraction diagnosis, and more specifically, to a system and method for automatically generating a tooth extraction diagnosis plan. Background Art

[0002] In the field of oral medicine, especially in the diagnosis process of complex tooth extraction surgeries, accurate image analysis and accurate diagnosis conclusions are crucial for ensuring patient safety and improving treatment effects. Traditional tooth extraction diagnosis mainly relies on doctors' experience and two-dimensional X-ray films. This method has certain limitations, especially when dealing with complex tooth positions, root morphologies, and their relationships with surrounding important structures (such as the inferior alveolar nerve canal or maxillary sinus). With the development of technology, cone beam computed tomography (CBCT) has been widely used in oral medicine due to its high-resolution three-dimensional imaging ability. However, how to efficiently and accurately extract useful information from these large amounts of three-dimensional data and make scientific and reasonable diagnostic decisions based on it remains a challenge.

[0003] Existing technical solutions often face the problem of insufficient feature extraction when processing CBCT images. For example, when analyzing teeth and their surrounding structures, traditional methods may not be able to comprehensively capture subtle anatomical features, such as the specific morphology of tooth roots, the degree of curvature, and important parameters such as bone density. This directly affects the accuracy of diagnosis. In addition, blurred boundaries are another significant problem. Due to the limited contrast between teeth and surrounding tissues, combined with possible noise interference, it is difficult to define the boundaries of teeth, which in turn affects the subsequent segmentation accuracy and feature quantification process. These problems limit the effectiveness of automated systems in actual clinical applications and lead to a decrease in the reliability of diagnostic results.

[0004] Therefore, an optimized automatic generation solution for tooth extraction diagnosis plans is expected. Summary of the Invention

[0005] In view of the fact that traditional tooth extraction diagnosis methods rely on two-dimensional X-ray films, which have problems such as image overlap and loss of detailed information, and the manual analysis of CBCT image data is inefficient and subjective, making it difficult to accurately quantify tooth feature parameters and evaluate tooth extraction risks. This application proposes a system and method for automatically generating a tooth extraction diagnosis plan.

[0006] According to one aspect of the present application, an automatic generation system for an extraction diagnosis plan is provided, including: a CBCT image data acquisition module for acquiring CBCT image data of a patient object; an image preprocessing module for preprocessing the CBCT image data to obtain preprocessed CBCT image data; a dental image segmentation module for performing refined segmentation based on boundary enhancement on the preprocessed CBCT image data to obtain a dental image segmentation result map, wherein, in the process of refined segmentation based on boundary enhancement, a direction sensitivity modulation map is constructed to perform regional boundary saliency; a dental feature quantification and extraction module for performing multi-dimensional feature extraction and quantification based on the dental image segmentation result map to obtain dental quantification feature parameters; an extraction intelligent diagnosis module for inputting the dental quantification feature parameters into a pre-constructed clinical knowledge rule base to obtain an extraction intelligent diagnosis conclusion; a structured diagnosis report generation module for integrating the dental image segmentation result map, the dental quantification feature parameters, and the extraction intelligent diagnosis conclusion to obtain a structured diagnosis report.

[0007] In the above automatic generation system for an extraction diagnosis plan, the image preprocessing module is used to: standardize, denoise, and enhance the contrast of the CBCT image data to obtain the preprocessed CBCT image data.

[0008] In the above automatic generation system for an extraction diagnosis plan, the dental image segmentation module includes: a dental image basic segmentation unit for inputting the preprocessed CBCT image data into a basic segmentation network to obtain a dental image local visual feature map and a dental image global visual feature map; a dental image multi-level fusion unit for fusing the dental image local visual feature map and the dental image global visual feature map to obtain a dental image multi-level visual feature map; a dental image boundary saliency prediction unit for inputting the dental image multi-level visual feature map into a boundary saliency prediction network to obtain a boundary saliency prediction feature map; a dental image boundary enhancement unit for fusing the dental image multi-level visual feature map and the boundary saliency prediction feature map to obtain a dental image boundary enhanced visual feature map; a dental image segmentation prediction unit for inputting the dental image boundary enhanced visual feature map into a segmentation prediction layer to obtain the dental image segmentation result map.

[0009] In the above automatic generation system for an extraction diagnosis plan, the basic segmentation network includes a hybrid model integrating CNN and Transformer.

[0010] In the above automatic generation system for tooth extraction diagnosis schemes, the tooth image boundary saliency prediction unit is configured to: set a boundary segmentation threshold; mark positions where all pixel values in the multi-level visual feature map of the tooth image are greater than the boundary segmentation threshold as 1 and the remaining positions as 0 to obtain the boundary saliency prediction feature map.

[0011] In the above automatic generation system for tooth extraction diagnosis schemes, the tooth image boundary enhancement unit includes: a boundary saliency fusion subunit configured to fuse the multi-level visual feature map of the tooth image and the boundary saliency prediction feature map to obtain a tooth image boundary salient visual feature map; and a weighted calculation subunit configured to calculate the position-wise weighted sum between the multi-level visual feature map of the tooth image and the tooth image boundary salient visual feature map to obtain the tooth image boundary enhanced visual feature map.

[0012] In the above automatic generation system for tooth extraction diagnosis schemes, the boundary saliency fusion subunit is configured to: calculate the position-wise dot product between the multi-level visual feature map of the tooth image and the boundary saliency prediction feature map to obtain an initial tooth image boundary salient visual feature map; calculate the oriented gradient magnitude feature map of the multi-level visual feature map of the tooth image with respect to the boundary saliency prediction feature map; calculate the partial derivative of the feature values at each position in the oriented gradient magnitude feature map with respect to the feature values at each position in the multi-level visual feature map of the tooth image to obtain a first two-way partial derivative structure interaction coefficient modulation map; calculate the partial derivative of the feature values at each position in the oriented gradient magnitude feature map with respect to the feature values at each position in the boundary saliency prediction feature map to obtain a second two-way partial derivative structure interaction coefficient modulation map; construct a direction sensitivity modulation map based on the first two-way partial derivative structure interaction coefficient modulation map and the second two-way partial derivative structure interaction coefficient modulation map; and calculate the position-wise dot product between the direction sensitivity modulation map and the initial tooth image boundary salient visual feature map to obtain the tooth image boundary salient visual feature map.

[0013] In the above automatic generation system for tooth extraction diagnosis schemes, the tooth feature quantization and extraction module includes: a tooth structure and morphology measurement unit configured to perform structure and morphology measurement on the tooth image segmentation result map to obtain estimated values of the root length, diameter, curvature, alveolar bone height, and bone density; a tooth spatial relationship calculation unit configured to perform spatial relationship calculation on the tooth image segmentation result map to obtain the shortest distance between the root apex and the inferior alveolar nerve canal or the floor of the maxillary sinus; and a tooth lesion feature analysis unit configured to perform lesion feature analysis on the tooth image segmentation result map to obtain the volume, position, and gray-scale features of apical lesions, cysts, etc.

[0014] In the automatic generation system of the above-mentioned tooth extraction diagnosis scheme, the intelligent tooth extraction diagnosis module is used to: input the tooth quantization feature parameters into a pre-constructed clinical knowledge rule base to obtain a tooth extraction difficulty assessment result and a main risk prediction result, and the tooth extraction difficulty assessment result and the main risk prediction result constitute the intelligent tooth extraction diagnosis conclusion.

[0015] According to another aspect of the present application, there is also provided a method for automatically generating a tooth extraction diagnosis scheme. This method uses the above-mentioned automatic generation system of the tooth extraction diagnosis scheme. The method includes: obtaining CBCT image data of a patient object; performing image preprocessing on the CBCT image data to obtain preprocessed CBCT image data; performing refined segmentation based on boundary enhancement on the preprocessed CBCT image data to obtain a tooth image segmentation result map. Among them, during the refined segmentation based on boundary enhancement, a direction sensitivity modulation map is constructed to perform regional boundary saliency; performing multi-dimensional feature extraction and quantization based on the tooth image segmentation result map to obtain tooth quantization feature parameters; inputting the tooth quantization feature parameters into a pre-constructed clinical knowledge rule base to obtain an intelligent tooth extraction diagnosis conclusion; integrating the tooth image segmentation result map, the tooth quantization feature parameters, and the intelligent tooth extraction diagnosis conclusion to obtain a structured diagnosis report.

[0016] Compared with the prior art, the automatic generation system and method of the tooth extraction diagnosis scheme provided by the present application collect patient data through the CBCT image data acquisition module, and perform standardization, denoising, and contrast enhancement processing through the image preprocessing module to improve the image quality. Subsequently, the tooth image segmentation module performs refined segmentation based on boundary enhancement on the preprocessed image to accurately separate the tooth structure. The tooth feature quantization extraction module further extracts multi-dimensional feature parameters from the segmentation result, including key indicators such as root length and curvature. Finally, the intelligent tooth extraction diagnosis module combines the clinical knowledge rule base and automatically gives a tooth extraction difficulty assessment and a main risk prediction conclusion based on the extracted feature parameters. The structured diagnosis report generation module integrates all relevant information to form a comprehensive diagnosis recommendation, thereby significantly improving the accuracy and efficiency of tooth extraction diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] By describing the examples of the present application in more detail in conjunction with the drawings, the above-mentioned and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the examples of the present application, and constitute a part of the specification. Together with the examples of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 FIG. shows a schematic block diagram of an automatic generation system of a tooth extraction diagnosis scheme according to an example of the present application.

[0019] Figure 2 The schematic block diagram of the tooth image segmentation module in the automatic generation system of the tooth extraction diagnosis scheme according to an example of the present application is illustrated.

[0020] Figure 3 The schematic block diagram of the tooth image boundary enhancement unit in the automatic generation system of the tooth extraction diagnosis scheme according to an example of the present application is illustrated.

[0021] Figure 4 The schematic block diagram of the tooth feature quantification and extraction module in the automatic generation system of the tooth extraction diagnosis scheme according to an example of the present application is illustrated.

[0022] Figure 5 The schematic flowchart of the automatic generation method of the tooth extraction diagnosis scheme according to an example of the present application is illustrated. Detailed implementation manners

[0023] Next, examples of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described examples are only a part of the examples of the present application, rather than all the examples of the present application. It should be understood that the present application is not limited by the examples described herein.

[0024] Figure 1 The schematic flowchart of the automatic generation system of the tooth extraction diagnosis scheme according to an example of the present application is illustrated. As Figure 1 shown, the present application provides an automatic generation system 10 of a tooth extraction diagnosis scheme, including: a CBCT image data acquisition module 11, configured to acquire CBCT image data of a patient object; an image preprocessing module 12, configured to perform image preprocessing on the CBCT image data to obtain preprocessed CBCT image data; a tooth image segmentation module 13, configured to perform refined segmentation based on boundary enhancement on the preprocessed CBCT image data to obtain a tooth image segmentation result map, wherein, in the process of refined segmentation based on boundary enhancement, regional boundary saliency is performed by constructing a direction sensitivity modulation map; a tooth feature quantification and extraction module 14, configured to perform multi-dimensional feature extraction and quantification based on the tooth image segmentation result map to obtain tooth quantification feature parameters; a tooth extraction intelligent diagnosis module 15, configured to input the tooth quantification feature parameters into a pre-constructed clinical knowledge rule base to obtain a tooth extraction intelligent diagnosis conclusion; and a structured diagnosis report generation module 16, configured to integrate the tooth image segmentation result map, the tooth quantification feature parameters, and the tooth extraction intelligent diagnosis conclusion to obtain a structured diagnosis report.

[0025] Specifically, in the CBCT image data acquisition module 11, the CBCT image data of the patient object is acquired. Specifically, the process of acquiring the CBCT image data of the patient object is completed by an experienced radiographer operating a cone beam computed tomography (CBCT) device in a professional dental clinic or hospital. First, the patient needs to assume a suitable position according to the instructions, usually sitting on a special chair and fixing the head in a specific position to reduce artifacts caused by movement. Then, by adjusting the parameter settings of the CBCT device, such as voltage, current, and exposure time, according to the specific situation of the patient (such as age, body mass index, examination site, etc.), the image quality is optimized while minimizing the radiation dose.

[0026] In a specific example, when diagnosing a young adult who needs to have their mandibular wisdom teeth extracted, the technician will select an appropriate scanning range, focusing on covering the mandible and its surrounding structures, including the inferior alveolar nerve canal. During this process, the CBCT device rotates around the patient's head for one week, emitting X-rays during which the signals attenuated by different tissues are received by the detector. These raw data are then transmitted to the computer system. It should be noted that in order to ensure the integrity and accuracy of the image data, various interference factors in the scanning environment, such as electromagnetic interference and temperature changes, must be strictly controlled, which can be achieved through the use of shielding materials and technical means.

[0027] In a preferred example, considering the diversity and complexity in practical applications, acquiring the CBCT image data of the patient object also includes how to effectively manage and store these huge data sets. Specifically, on the one hand, with the development of digital technology, medical institutions can now use cloud storage solutions to securely store the CBCT image data of patients, which not only facilitates cross-departmental and even cross-border collaboration and communication, but also is convenient for long-term follow-up and observation of the disease progression. On the other hand, for scenarios with high real-time requirements, such as rapid assessment before emergency surgery, it may be necessary to deploy high-performance computing resources to accelerate the image reconstruction process to ensure that doctors can obtain clear and accurate diagnostic basis in the shortest time.

[0028] Specifically, in the image preprocessing module 12, the CBCT image data is preprocessed to obtain preprocessed CBCT image data. It should be understood that in the field of oral medicine, especially for tooth extraction diagnosis, obtaining high-quality CBCT image data is crucial. However, the original CBCT image data often has problems such as noise and low contrast. If these problems are not addressed, they will directly affect subsequent tooth image segmentation, feature extraction, and the final intelligent diagnosis results. Therefore, in one example, the image preprocessing module 12 is used to: standardize, denoise, and enhance the contrast of the CBCT image data to obtain the preprocessed CBCT image data.

[0029] Specifically, the standardization process is to ensure the consistency of CBCT image data collected from different sources and under different conditions. Since images taken under different devices and different setting parameters may have differences in aspects such as brightness and grayscale, if these differences are not eliminated, it will lead to inaccurate subsequent analysis results. Standardization adjusts the grayscale range of the image to conform to a predetermined standard interval, such as between 0 and 255. The purpose of this is to ensure that all images input into the system are on the same scale, thus facilitating unified processing.

[0030] At the same time, denoising is one of the key steps to improve image quality. The noise in CBCT image data may come from various factors, such as the electronic noise of the device itself and artifacts caused by patient movement. Removing this noise not only helps to improve the clarity of the image but also reduces the possible misjudgments in the subsequent processing stage. Common denoising methods include Gaussian filtering, median filtering, etc. For example, using Gaussian filtering can smooth out some random noise while retaining important details; while median filtering is particularly suitable for removing salt-and-pepper noise, which achieves the denoising effect by replacing the central pixel value with the median value of the neighboring pixel values.

[0031] Contrast enhancement is crucial for highlighting the key structures in the image. Good contrast can make it easier for doctors or automated systems to distinguish different tissue types, such as the boundary between teeth and surrounding bone tissue. Common contrast enhancement techniques include histogram equalization, adaptive contrast enhancement, etc. Histogram equalization expands the dynamic range by redistributing the grayscale values of the image, thus making the details in the image more clearly visible; adaptive contrast enhancement adjusts the contrast according to the characteristics of the local area, which is especially suitable for processing images with large brightness variations.

[0032] In a specific example, after the CBCT image data of the patient is imported into the system, it is first converted into a standard format through a normalization algorithm to ensure that all subsequent operations are based on a consistent basis. Then, a denoising algorithm is applied to remove various possible interference factors, such as using a Gaussian filter to smooth the image and reduce unnecessary noise. Subsequently, histogram equalization or adaptive contrast enhancement techniques are adopted to improve the overall contrast of the image, making the teeth and their surrounding structures more distinct, facilitating accurate segmentation and feature extraction by subsequent modules. Through such a series of meticulous preprocessing steps, not only can the quality of the image be significantly improved, but also more reliable data support can be provided for subsequent intelligent diagnosis, thereby enhancing the accuracy and efficiency of the entire tooth extraction diagnosis scheme generation system.

[0033] Specifically, in the tooth image segmentation module 13, refined segmentation based on boundary enhancement is performed on the preprocessed CBCT image data to obtain a tooth image segmentation result map. It should be understood that traditional single models often struggle to meet the requirements of high-precision segmentation simultaneously. For example, relying solely on CNN may not fully consider global information, while using Transformer alone may lead to excessive consumption of computing resources. In contrast, a hybrid model integrating CNN and Transformer can efficiently extract local details while taking into account global structural information, making it very suitable for segmentation tasks of objects with complex geometric characteristics such as teeth. In addition, through strategies such as multi-level feature fusion and boundary saliency enhancement, the quality of the segmentation result can be significantly improved, reducing the misjudgment rate and providing more reliable support for subsequent diagnosis.

[0034] In an example, as Figure 2 shown, the tooth image segmentation module 13 includes: a tooth image basic segmentation unit 131 for inputting the preprocessed CBCT image data into a basic segmentation network to obtain a tooth image local visual feature map and a tooth image global visual feature map; a tooth image multi-level fusion unit 132 for fusing the tooth image local visual feature map and the tooth image global visual feature map to obtain a tooth image multi-level visual feature map; a tooth image boundary saliency prediction unit 133 for inputting the tooth image multi-level visual feature map into a boundary saliency prediction network to obtain a boundary saliency prediction feature map; a tooth image boundary enhancement unit 134 for fusing the tooth image multi-level visual feature map and the boundary saliency prediction feature map to obtain a tooth image boundary enhanced visual feature map; a tooth image segmentation prediction unit 135 for inputting the tooth image boundary enhanced visual feature map into a segmentation prediction layer to obtain the tooth image segmentation result map.

[0035] Specifically, first, the preprocessed CBCT image data is input into the basic segmentation network. In one example, the basic segmentation network includes a hybrid model integrating CNN and Transformer. Specifically, CNN is good at capturing local detail information in images, which is crucial for accurately identifying the specific shape, size, and position of teeth. On the other hand, Transformer is known for its powerful global context understanding ability and can effectively model long-range dependencies in the entire image, helping to more precisely define the boundaries between teeth and other structures. In a specific example, the data representation of the input image is initialized through a multi-layer perceptron (MLP) embedding layer, and then the feature representation is gradually refined through a series of alternately stacked CNN blocks and Transformer encoder layers.

[0036] Here, it can be understood by those skilled in the art that the CNN block usually consists of a series of convolutional layers, activation functions, and pooling layers. The convolutional layer is the core part of the CNN. It uses a set of learnable filters (or kernels) to slide on the input data to detect different local features, such as edges, textures, etc. These filters are trained through the backpropagation algorithm to gradually optimize their parameters for more accurately identifying specific patterns. Activation functions, such as ReLU (Rectified Linear Unit), are applied after each convolutional layer to introduce non-linearity, which is crucial for modeling complex input-output relationships. In addition, the pooling layer, especially the max pooling layer, is often used to reduce the spatial size of the feature map, thereby reducing the computational complexity and controlling the risk of overfitting. Through the above design, the CNN block can efficiently compress information without losing key details, providing a concise but informative feature representation for subsequent processing. On the other hand, the Transformer encoder layer adopts a completely different mechanism to process sequence data, especially those tasks that require understanding global context. A core concept of the Transformer architecture is the self-attention mechanism, which allows the model to consider the information of all other elements for each element in the sequence. This means that for the given dental CBCT image data, the Transformer can dynamically adjust the weights between different regions to ensure that even structures far from the current focus can be appropriately attended to. This feature makes the Transformer perform well when dealing with objects with complex geometric shapes because it can not only capture local details but also understand the macroscopic structure of the entire image. Specifically, the Transformer encoder layer consists of two main parts: the multi-head self-attention mechanism and the feed-forward neural network. The former is responsible for generating query, key, and value matrices and determining the importance of each position by calculating the similarity scores between them; the latter further processes these weighted feature representations to enhance the learning ability of the model.

[0037] Next, after obtaining the local and global visual feature maps, it is necessary to fuse the two to form a multi-level visual feature map. Here, a method based on the attention mechanism is adopted, enabling the model to dynamically adjust its focus at different scales, thereby better capturing the subtle structural changes of the teeth. In a specific example, by calculating the similarity matrix between the two feature maps and using the softmax function to convert it into a weight matrix, and then performing weighted summation to obtain the final multi-level visual feature map. This method allows the model to automatically focus on the most informative parts, improving the accuracy of segmentation.

[0038] Then, in order to further clarify the boundaries of the teeth, a boundary saliency prediction network is used to generate a boundary saliency prediction feature map. In an example, the tooth image boundary saliency prediction unit 133 is used to: set a boundary segmentation threshold, mark the positions where all pixel values in the multi-level visual feature map of the tooth image are greater than the boundary segmentation threshold as 1 and the remaining positions as 0 to obtain the boundary saliency prediction feature map. It should be understood that the boundary segmentation threshold is determined based on existing medical knowledge and empirical data. For example, in common CBCT images, teeth (especially enamel) may have relatively high gray values, approximately between 1500 and 3000 Hounsfield units (HU); while the gray values of bone tissue are relatively low, but still significantly higher than those of soft tissues, roughly in the range of 500 to 1200 HU. Based on this, the present application selects a fixed threshold between the two for the segmentation operation. In a specific example, it is decided to adopt 1200 HU as the boundary segmentation threshold. This means that in the subsequent processing, for each pixel point in the multi-level visual feature map, if its gray value is greater than 1200 HU, it is considered that the pixel belongs to the tooth region and is marked as 1; conversely, if the gray value is less than or equal to 1200 HU, it is regarded as the background (i.e., non-tooth region) and is marked as 0. Of course, the above is only an example and is not specifically limited in this example.

[0039] However, directly applying a fixed threshold may lead to problems such as blurred or discontinuous edges. Therefore, in a preferred example, an adaptive threshold adjustment strategy is introduced, such as the Otsu algorithm, which can automatically select the optimal threshold based on the image histogram, thereby ensuring the clarity and coherence of the boundaries. Specifically, first, collect the grayscale values of all pixels to be processed and construct its histogram. This histogram reflects the distribution of the number of pixels at different grayscale levels. Next, apply the Otsu algorithm to calculate the optimal threshold. The core idea of this algorithm is to find a threshold that maximizes the between-class variance between the foreground and the background. In other words, it is necessary to find a grayscale level such that the two parts (foreground and background) divided by this boundary are as separated from each other as possible and as consistent as possible internally. Specifically, the Otsu algorithm will traverse all possible thresholds. For each threshold, calculate the average grayscale values of the foreground and the background, and calculate the between-class variance based on this. Finally, select the threshold that maximizes the between-class variance as the boundary segmentation threshold.

[0040] Finally, the tooth image boundary is strengthened by fusing multi-level visual feature maps and boundary saliency prediction feature maps. In an example, as Figure 3 shown, the tooth image boundary strengthening unit 134 includes: a boundary saliency fusion subunit 1341 for fusing the multi-level visual feature map of the tooth image and the boundary saliency prediction feature map to obtain a tooth image boundary salient visual feature map; a weighted calculation subunit 1342 for calculating the position-wise weighted sum between the multi-level visual feature map of the tooth image and the tooth image boundary salient visual feature map to obtain the tooth image boundary strengthened visual feature map.

[0041] Here, considering that when fusing the multi-level visual feature map of the tooth image and the boundary saliency prediction feature map, first perform a position-wise dot multiplication on the multi-level visual feature map of the tooth image and the boundary saliency prediction feature map to obtain an initial tooth image boundary salient visual feature map for boundary saliency prediction of the multi-level visual features of the tooth image data. However, obviously, in the visual feature representation of the actual multi-level visual feature map of the tooth image, there may be obvious extreme value distribution characteristics of the gradient magnitude when the visually significant regions cross the boundary segmentation, thereby affecting the adaptive sensitivity of the region significant edges on the overall feature map.

[0042] Based on this, represent the multi-level visual feature map of the tooth image as , and represent the boundary saliency prediction feature map as . First, calculate the directional gradient magnitude feature map of the multi-level visual feature map of the tooth image relative to the boundary saliency prediction feature map, denoted as: ; Among them, represents subtraction by position, represents multiplication by position, represents the oriented gradient magnitude feature map.

[0043] That is, through the above formula, the multi-level visual feature map of the tooth image with respect to the boundary saliency prediction feature map is defined by the oriented gradient magnitude, so that the direction distribution of the gradient field is used as the core constraint condition for saliency. Then, for the extreme value distribution, its partial derivatives are calculated respectively. Specifically, the partial derivatives of the feature values at each position in the oriented gradient magnitude feature map with respect to the feature values at each position in the multi-level visual feature map of the tooth image are calculated to obtain the first two-way partial derivative structure interaction coefficient modulation map, and the partial derivatives of the feature values at each position in the oriented gradient magnitude feature map with respect to the feature values at each position in the boundary saliency prediction feature map are calculated to obtain the second two-way partial derivative structure interaction coefficient modulation map, which is expressed as: ; ; where represents calculating the reciprocal of the feature value at each position in the multi-level visual feature map of the tooth image, represents calculating the square of the reciprocal of the feature value at each position in the multi-level visual feature map of the tooth image.

[0044] That is, by constructing a two-way partial derivative structure interaction, the gradient information of the gradient field is associated with the cross-boundary segmentation distribution of the multi-level visual features in a differentiable extreme value sampling manner.

[0045] Then, based on the first two-way partial derivative structure interaction coefficient modulation map and the second two-way partial derivative structure interaction coefficient modulation map, a direction sensitivity modulation map is constructed, which is expressed as: ; where represents a learnable hyperparameter, represents addition by position, represents the direction sensitivity modulation map.

[0046] By constructing the direction sensitivity modulation map, regional boundary saliency is performed, that is, adaptive normalization is performed for the direction sensitivity. Here, the learnable hyperparameter is introduced to make the boundary direction information consistent with the gradient information.

[0047] Finally, the significant visual feature map of the tooth image boundary is obtained by performing a point - by - point multiplication between the direction - sensitivity modulation map and the significant visual feature map of the initial tooth image boundary. In this way, when the multi - level visual feature map of the tooth image is fused with the significant visual feature map of the tooth image boundary through point - addition, the adaptive significant - sensitive consistency between significant distinguishable features can be achieved based on the gradient direction sensitivity, thereby enhancing the expression effect of the enhanced visual feature map of the tooth image boundary.

[0048] On this basis, the present application also uses a segmentation prediction layer to perform a final segmentation operation on the enhanced feature map. The segmentation prediction layer usually consists of several convolutional layers, and its function is to perform pixel - by - pixel classification on the enhanced visual feature map of the tooth image boundary to determine whether each pixel belongs to the background or the target region, so as to obtain the tooth image segmentation result map. In one example, inputting the enhanced visual feature map of the tooth image boundary into the segmentation prediction layer to obtain the tooth image segmentation result map includes: First, a series of 1x1 convolutional kernels are used to reduce the number of channels of the feature map, and the feature map is adjusted through this convolutional operation to prepare for classification. The 1x1 convolution plays a role in dimensionality reduction and feature recombination here, which helps to simplify the subsequent calculation process. Then, the Softmax function is used to classify each pixel. Softmax assigns a probability value belonging to the target class to each pixel, so that the probability distribution of each pixel point belonging to the tooth can be obtained. The class with the highest probability is selected as the label of the pixel point, thus completing the segmentation. After the above - mentioned processing is completed, the segmentation prediction layer directly outputs the final tooth image segmentation result map, which clearly defines the boundary between the tooth and the background region, providing an accurate basis for the subsequent quantitative extraction of tooth features.

[0049] Specifically, in the tooth feature quantification extraction module 14, multi - dimensional feature extraction and quantification are performed based on the tooth image segmentation result map to obtain tooth quantification feature parameters. It should be understood that the specific conditions of each patient are different. Therefore, personalized treatment plans are particularly important. Multi - dimensional feature extraction and quantification can reveal the differences between individuals, thus supporting more accurate diagnosis and treatment decisions. For example, for patients suffering from specific diseases (such as osteoporosis), their bone density may be lower than the normal range, which requires adjusting the surgical plan to ensure safety and effectiveness. By comprehensively analyzing the tooth and its surrounding structures, doctors can customize the most suitable treatment strategy for each patient's specific situation, improving the treatment effect while reducing risks.

[0050] In one example, such as Figure 4As shown, the tooth feature quantification and extraction module 14 includes: a tooth structure and morphology measurement unit 141 for measuring the structure and morphology of the tooth image segmentation result map to obtain estimated values of the root length, diameter, curvature, alveolar bone height, and bone density; a tooth spatial relationship calculation unit 142 for calculating the spatial relationship of the tooth image segmentation result map to obtain the shortest distance between the root apex and the inferior alveolar nerve canal or the floor of the maxillary sinus; and a tooth lesion feature analysis unit 143 for analyzing the lesion features of the tooth image segmentation result map to obtain the volume, location, and gray-scale features of apical lesions, cysts, etc.

[0051] In a specific example, first, when measuring the tooth structure and morphology, geometric calculation methods can be used to obtain parameters such as the root length, diameter, and curvature. For example, for the measurement of the root length, two endpoints on the tooth contour (i.e., the top of the crown and the end of the root) can be detected, and then the Euclidean distance between these two points can be calculated. This method is simple and intuitive and applicable to most cases. To improve accuracy, edge detection techniques such as the Canny detector can be combined. First, accurately locate the tooth contour, and then calculate the length based on the feature points on the contour. Similarly, the root diameter can be estimated by measuring the maximum and minimum diameters on a specific cross-section, and the curvature can be approximated by fitting a curve to represent the curved shape of the tooth, and the curvature of this curve can be calculated as a measure. These geometric calculations are not only easy to implement but also can quickly provide preliminary measurement results, laying a foundation for further refinement.

[0052] Secondly, in terms of calculating the tooth spatial relationship, especially when determining the distance between the root apex and important anatomical structures (such as the inferior alveolar nerve canal or the floor of the maxillary sinus), an accurate spatial positioning system is required. Here, three-dimensional reconstruction technology can be considered to convert the two-dimensional segmentation result map into a three-dimensional model. Based on this three-dimensional model, the shortest Euclidean distance between these points of interest can be found through the nearest neighbor search algorithm. For example, first define the specific position coordinates of the inferior alveolar nerve canal or the floor of the maxillary sinus, and then use the distance formula in three-dimensional space to calculate the straight-line distance between these points and the root apex. In addition, to improve the calculation efficiency, a small lookup table or hash table can be pre-trained specifically for quickly finding the nearest neighbor points, which can not only ensure the calculation speed but also maintain a high accuracy level. This method based on geometry and mathematics can efficiently solve practical problems without relying on complex models, demonstrating its practicality and flexibility.

[0053] Finally, in the analysis of dental lesion characteristics, it involves the identification and quantification of pathological features such as apical lesions and cysts. This step usually requires more meticulous analysis methods. A feasible solution is to use traditional image processing techniques, such as region growing algorithms or watershed algorithms, to segment the lesion area. Then, estimate its volume size by calculating the number of pixels in the lesion area; at the same time, use color space conversion technology to convert the original RGB image to a color space more suitable for analyzing lesion characteristics (such as HSV), and then extract the color and grayscale features of the lesion site. In addition, threshold segmentation technology can also be introduced to distinguish the lesion area from other normal tissues according to the specific grayscale range of the lesion area. Although this method is relatively simple, it performs well when dealing with specific types of lesions, especially in scenarios where the data volume is not large or large-scale automation is not required, and its advantages are more obvious.

[0054] Furthermore, considering that traditional manual measurement methods are difficult to handle complex dental morphologies and lesion conditions, and a single automated tool may perform poorly on certain specific tasks. In contrast, this application also proposes a deep learning model with multi-level and multi-modal fusion, which can make full use of the advantages of modern computer vision technology to achieve efficient and accurate extraction of dental features. At the same time, by integrating a variety of advanced technical means, such as three-dimensional reconstruction, nearest neighbor search, and transfer learning, etc., not only the functional diversity of the system is enhanced, but also the stability and reliability of the overall performance are improved.

[0055] In a preferred example, first, when measuring the dental structure morphology, a method combining convolutional neural network (CNN) and traditional image processing techniques can be adopted. For example, use the U-Net architecture as the basic model, which is widely used in the field of medical image analysis due to its excellent semantic segmentation ability. U-Net can effectively capture the boundary information of the teeth and generate high-quality segmentation result maps. Based on this segmentation result map, further use traditional edge detection algorithms such as the Canny detector to accurately locate the tooth contour, and then apply geometric calculation methods to measure parameters such as the root length, diameter, and curvature of the tooth. In addition, three-dimensional reconstruction technology can also be introduced to construct a three-dimensional model of the tooth according to the two-dimensional segmentation result map, so as to more accurately estimate indicators such as the alveolar bone height and bone density. The advantage of this method is that it not only utilizes the powerful feature extraction ability of the deep learning model, but also combines the high efficiency of traditional algorithms in specific tasks, ensuring the accuracy of the measurement results.

[0056] Secondly, for the calculation of tooth spatial relationships, especially the measurement of the distance between the root apex and important anatomical structures (such as the inferior alveolar nerve canal or the floor of the maxillary sinus), an accurate spatial positioning system is required. Here, a model based on the Transformer architecture can be considered to enhance the global context understanding ability, because the Transformer is good at dealing with long-range dependencies, which is particularly important for cross-regional spatial distance calculation. Specifically, a Transformer encoder layer can be added on the basis of the above U-Net to capture the global information in the entire image. Then, by defining specific points of interest (POIs), such as the positions of the inferior alveolar nerve canal or the floor of the maxillary sinus, the shortest Euclidean distance between these POIs and the root apex is found using the nearest neighbor search algorithm. To improve the calculation efficiency, a small neural network can also be pre-trained specifically for quickly finding the nearest neighbor, which can not only ensure the calculation speed but also maintain a high accuracy level.

[0057] Finally, in terms of the analysis of tooth lesion characteristics, it involves the identification and quantification of pathological features such as apical lesions and cysts. In the specific example of this application, the Mask R-CNN framework is adopted, which is a powerful tool designed specifically for instance segmentation tasks. Mask R-CNN can not only accurately locate the lesion area but also generate an accurate mask for it, facilitating subsequent quantitative analysis. On this basis, the volume size can be estimated by calculating the number of pixels in the lesion area; at the same time, the original RGB image is converted to a color space more suitable for analyzing lesion characteristics (such as HSV) using color space conversion technology, and then the color and gray-scale characteristics of the lesion site are extracted. In addition, a transfer learning strategy can be introduced, that is, the network is initialized with the weights of a pre-trained model and then fine-tuned for specific lesion types. This method not only accelerates the training process but also improves the adaptability and generalization ability of the model to different cases.

[0058] Specifically, in the tooth extraction intelligent diagnosis module 15, the tooth quantization feature parameters are input into a pre-constructed clinical knowledge rule base to obtain a tooth extraction intelligent diagnosis conclusion. It should be understood that traditional diagnostic methods often rely on the experience and technical level of individual doctors and are easily affected by subjective factors. The method based on the clinical knowledge rule base proposed in this application attempts to overcome these problems by integrating multi-source data and expert wisdom to establish an objective and systematic decision support framework. On the one hand, this approach helps to standardize the diagnosis and treatment process and reduce the probability of human errors; on the other hand, with the help of advanced data analysis techniques and automated tools, the work efficiency can be significantly improved, enabling even inexperienced medical staff to make reasonable treatment choices. In addition, considering the dynamics and diversity of the medical environment, the design of the rule base should also have good scalability and flexibility to absorb new research results and practical experiences at any time.

[0059] In one example, the intelligent tooth extraction diagnosis module 15 is configured to: input the tooth quantization feature parameters into a pre-constructed clinical knowledge rule base to obtain a tooth extraction difficulty assessment result and a main risk prediction result, and the tooth extraction difficulty assessment result and the main risk prediction result constitute the intelligent tooth extraction diagnosis conclusion.

[0060] In a specific example, constructing a clinical knowledge rule base includes: First, collect data from multiple sources, including but not limited to electronic health records (EHRs), clinical trial results, professional literature, and feedback information in actual operations, etc. For example, in the initial stage of construction, data mining techniques can be used to extract tooth extraction-related case materials from existing EHR databases, and these materials cover information in multiple dimensions such as patient age, gender, tooth condition, and past medical history. Then, invite experts in the field to participate in the design of the rule base. They define a series of rules based on their clinical experience and the latest research results to guide diagnostic decisions under different feature combinations. For example, for specific types of tooth lesions or anatomical structure abnormalities, corresponding treatment suggestions and expected risk levels are specified. In addition, machine learning algorithms can be used to optimize and adjust the above rules to ensure that the rule base can adapt to the changing needs of clinical practice.

[0061] In practical applications, when the tooth quantization feature parameters are extracted, they will be passed as inputs to this knowledge rule base. The key here lies in how to effectively map these parameters to specific diagnostic conclusions. For this purpose, a method based on fuzzy logic reasoning can be adopted. This method allows handling uncertainty and imprecision and is very suitable for medical decision support systems. For example, assume that a certain patient has a relatively severe root curvature and a low estimated bone density value. Then, according to the relevant rules in the rule base, the system may infer a high tooth extraction difficulty score and predict possible main risks such as postoperative infection or bleeding. At the same time, to improve the reasoning efficiency, an indexing mechanism or a hash table can be introduced into the rule base to quickly locate the best rule set applicable to the current feature combination.

[0062] Specifically, in the structured diagnosis report generation module 16, the tooth image segmentation result map, the tooth quantification feature parameters, and the intelligent tooth extraction diagnosis conclusion are integrated to obtain a structured diagnosis report. It should be understood that traditional unstructured reports often lack a unified standard, which easily leads to information omission or misinterpretation. However, this application solves this problem by integrating the tooth image segmentation result map, the tooth quantification feature parameters, and the intelligent tooth extraction diagnosis conclusion. On the one hand, by using a standardized template, it can ensure that the format of each generated report is consistent, facilitating doctors to quickly obtain the required information. On the other hand, with the help of modern information technologies, such as the application of image processing technology and database management systems, not only can the work efficiency be improved, but also the authenticity and integrity of the data can be guaranteed.

[0063] In a specific example, the structured diagnosis report includes the following main parts: 1. Patient basic information: This includes basic information such as the patient's name, age, gender, and medical record number. This part of the content is mainly used to identify the object to which the report belongs and ensure the accuracy of the information. 2. Display of tooth image segmentation result maps: In this part, the tooth image segmentation result maps are embedded in the report. These images should clearly mark the specific positions of the teeth and the important anatomical structures around them. To facilitate doctors' viewing, interactive functions, such as zoom in / out and rotation operation options, can also be added, enabling doctors to observe the teeth and their surrounding conditions from multiple angles. In addition, for areas of special concern, such as parts where lesions may exist, they can be highlighted, such as by highlighting, to help doctors quickly locate the key points. 3. List of tooth quantification feature parameters: List various tooth quantification feature parameters obtained through precise calculations, such as root length, diameter, curvature, alveolar bone height, and estimated bone density values. To enhance the understandability of the information, in addition to giving specific numerical values, reference values within the normal range should also be attached for comparison, making abnormal situations obvious at a glance. 4. Intelligent tooth extraction diagnosis conclusion: The intelligent tooth extraction diagnosis conclusion based on the clinical knowledge rule base should be listed in detail, including but not limited to the tooth extraction difficulty assessment result and the main risk prediction result. 5. Additional remarks and suggestions: Finally, space is reserved for doctors to add their personal opinions or other matters that need attention. This not only helps in formulating personalized treatment plans but also provides a basis for subsequent follow-up examinations.

[0064] In summary, the automatic generation system for tooth extraction diagnosis solutions provided in this application collects patient data through the CBCT image data acquisition module, and performs standardization, denoising, and contrast enhancement processing through the image preprocessing module to improve the image quality. Subsequently, the tooth image segmentation module performs refined segmentation based on boundary enhancement on the preprocessed image to accurately separate the tooth structure. The tooth feature quantification and extraction module further extracts multi-dimensional feature parameters from the segmentation results, including key indicators such as root length and curvature. Finally, the tooth extraction intelligent diagnosis module combines the clinical knowledge rule base and automatically gives the tooth extraction difficulty assessment and main risk prediction conclusions based on the extracted feature parameters. The structured diagnosis report generation module integrates all relevant information to form a comprehensive diagnosis recommendation, thereby significantly improving the accuracy and efficiency of tooth extraction diagnosis.

[0065] This application also provides an automatic generation method for tooth extraction diagnosis solutions. As Figure 5 shown, the automatic generation method for tooth extraction diagnosis solutions includes: S1, obtaining CBCT image data of a patient object; S2, performing image preprocessing on the CBCT image data to obtain preprocessed CBCT image data; S3, performing refined segmentation based on boundary enhancement on the preprocessed CBCT image data to obtain a tooth image segmentation result map, wherein, during the process of refined segmentation based on boundary enhancement, regional boundary saliency is performed by constructing a direction sensitivity modulation map; S4, performing multi-dimensional feature extraction and quantification based on the tooth image segmentation result map to obtain tooth quantification feature parameters; S5, inputting the tooth quantification feature parameters into a pre-constructed clinical knowledge rule base to obtain a tooth extraction intelligent diagnosis conclusion; S6, integrating the tooth image segmentation result map, the tooth quantification feature parameters, and the tooth extraction intelligent diagnosis conclusion to obtain a structured diagnosis report.

[0066] The basic principles of this application have been described above in combination with specific examples. However, it should be noted that the advantages, benefits, effects, etc. mentioned in this application are only examples and not limitations. It cannot be considered that these advantages, benefits, effects, etc. are essential for each example of this application. Additionally, the above-disclosed specific details are only for the purposes of illustration and facilitating understanding, and are not limitations. The above details do not limit this application to necessarily adopt the above specific details for implementation.

[0067] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any way. Words such as "comprising," "including," "having," etc. are open-ended terms meaning "including but not limited to" and can be used interchangeably with each other. The words "or" and "and" as used herein refer to the phrase "and / or" and can be used interchangeably with it, unless the context clearly indicates otherwise. The phrase "such as" as used herein refers to the phrase "such as but not limited to" and can be used interchangeably with it.

[0068] It should also be noted that in the devices, equipment, and methods of this application, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent solutions of this application.

[0069] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

[0070] The above description has been given for purposes of illustration and description. In addition, this description is not intended to limit the examples of this application to the forms disclosed herein. Although multiple example aspects and examples have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions, and sub-combinations thereof.

Claims

1. A system for automatically generating tooth extraction diagnosis plans, characterized in that: include: A CBCT image data acquisition module, used to acquire CBCT image data of a patient object; An image preprocessing module, used for performing image preprocessing on the CBCT image data to obtain preprocessed CBCT image data; A tooth image segmentation module, used for performing fine segmentation based on boundary enhancement on the pre-processed CBCT image data to obtain a tooth image segmentation result map, wherein in the process of fine segmentation based on boundary enhancement, a directional sensitivity modulation map is constructed to perform regional boundary salient; A tooth feature quantification extraction module, used for performing multi-dimensional feature extraction and quantification based on the tooth image segmentation result map to obtain tooth quantification feature parameters; A tooth extraction intelligent diagnosis module, used for inputting the tooth quantitative characteristic parameters into a pre-built clinical knowledge rule base to obtain a tooth extraction intelligent diagnosis conclusion; The structured diagnosis report generation module is used to integrate the tooth image segmentation result map, the tooth quantitative characteristic parameters and the tooth extraction intelligent diagnosis conclusion to obtain a structured diagnosis report.

2. The automatic generation system of tooth extraction diagnosis plan according to claim 1, characterized in that: The image preprocessing module is used to: perform standardization, denoising and contrast enhancement on the CBCT image data to obtain the preprocessed CBCT image data.

3. The automatic generation system of tooth extraction diagnosis plan according to claim 1, characterized in that: The tooth image segmentation module comprises: A dental image basic segmentation unit, used for inputting the pre-processed CBCT image data into a basic segmentation network to obtain a dental image local visual feature map and a dental image global visual feature map; A tooth image multi-level fusion unit, used for fusing the tooth image local visual feature map and the tooth image global visual feature map to obtain a tooth image multi-level visual feature map; A tooth image boundary saliency prediction unit, used for inputting the multi-level visual feature map of the tooth image into a boundary saliency prediction network to obtain a boundary saliency prediction feature map; A tooth image boundary enhancement unit, used for fusing the tooth image multi-level visual feature map and the boundary significance prediction feature map to obtain a tooth image boundary enhancement visual feature map; The tooth image segmentation prediction unit is used to input the tooth image boundary enhancement visual feature map into the segmentation prediction layer to obtain the tooth image segmentation result map.

4. The automatic generation system of tooth extraction diagnosis plan according to claim 3, characterized in that: The basic segmentation network includes a hybrid model integrating CNN and Transformer.

5. The automatic generation system of tooth extraction diagnosis plan according to claim 3, characterized in that: The tooth image boundary saliency prediction unit is used to: Set the boundary segmentation threshold; All positions where the pixel values ​​in the multi-level visual feature map of the tooth image are greater than the boundary segmentation threshold are marked as 1, and the remaining positions are marked as 0 to obtain the boundary significance prediction feature map.

6. The automatic generation system of tooth extraction diagnosis plan according to claim 3, characterized in that: The tooth image boundary strengthening unit comprises: A boundary saliency fusion subunit, used for fusing the multi-level visual feature map of the tooth image and the boundary saliency prediction feature map to obtain a boundary saliency visual feature map of the tooth image; The weighted calculation subunit is used to calculate the position-weighted sum between the multi-level visual feature map of the tooth image and the significant visual feature map of the tooth image boundary to obtain the tooth image boundary reinforcement visual feature map.

7. The automatic generation system of tooth extraction diagnosis plan according to claim 6, characterized in that: The boundary saliency fusion subunit is used to: Calculate the multiplication of the multi-level visual feature map of the tooth image and the boundary saliency prediction feature map by position point to obtain an initial tooth image boundary saliency visual feature map; Calculating a directional gradient amplitude feature map of the multi-level visual feature map of the tooth image relative to the boundary significance prediction feature map; Calculate the partial derivatives of the feature values ​​at each position in the directional gradient amplitude feature map relative to the feature values ​​at each position in the multi-level visual feature map of the tooth image to obtain a first bidirectional partial derivative structure interaction coefficient modulation map; Calculating partial derivatives of feature values ​​at each position in the directional gradient amplitude feature map relative to feature values ​​at each position in the boundary saliency prediction feature map to obtain a second bidirectional partial derivative structure interaction coefficient modulation map; Constructing a directional sensitivity modulation map based on the first bidirectional partial guide structure interaction coefficient modulation map and the second bidirectional partial guide structure interaction coefficient modulation map; The directional sensitivity modulation map and the initial tooth image boundary significant visual feature map are multiplied by the position points to obtain the tooth image boundary significant visual feature map.

8. The automatic generation system of tooth extraction diagnosis plan according to claim 1, characterized in that: The tooth feature quantitative extraction module comprises: A tooth structure morphology measurement unit, used for performing structure morphology measurement on the tooth image segmentation result image to obtain root length, diameter, curvature, alveolar bone height and bone density estimation values; A tooth spatial relationship calculation unit, used for performing spatial relationship calculation on the tooth image segmentation result map to obtain the shortest distance between the tooth root tip and the inferior alveolar nerve canal or the maxillary sinus floor; The tooth lesion feature analysis unit is used to perform lesion feature analysis on the tooth image segmentation result map to obtain the volume, position and grayscale features of the apical lesion, cyst and the like.

9. The automatic generation system of tooth extraction diagnosis plan according to claim 1, characterized in that: The tooth extraction intelligent diagnosis module is used for: The tooth quantitative characteristic parameters are input into a pre-constructed clinical knowledge rule base to obtain a tooth extraction difficulty assessment result and a major risk prediction result, and the tooth extraction difficulty assessment result and the major risk prediction result constitute the tooth extraction intelligent diagnosis conclusion.

10. A method for automatically generating a tooth extraction diagnosis plan, using the automatic generation system for a tooth extraction diagnosis plan according to any one of claims 1 to 9, characterized in that: include: Acquire CBCT image data of a patient object; Performing image preprocessing on the CBCT image data to obtain preprocessed CBCT image data; Performing refined segmentation based on boundary enhancement on the preprocessed CBCT image data to obtain a tooth image segmentation result map, wherein in the process of refined segmentation based on boundary enhancement, regional boundaries are marked by constructing a directional sensitivity modulation map; Perform multi-dimensional feature extraction and quantification based on the tooth image segmentation result map to obtain tooth quantitative feature parameters; Inputting the tooth quantitative characteristic parameters into a pre-built clinical knowledge rule base to obtain an intelligent diagnosis conclusion for tooth extraction; The tooth image segmentation result map, the tooth quantitative characteristic parameters and the tooth extraction intelligent diagnosis conclusion are integrated to obtain a structured diagnosis report.

Images