Second mesial buccal root canal measurement method and device and storage medium

Abstract

The invention discloses a second mesial buccal root canal measurement method and device and a storage medium, and belongs to the technical field of medical image processing. The method comprises the steps that an oral cavity CBCT image of a detected object is acquired, the position of a target molar is acquired from the oral cavity CBCT image through a deep learning method, a three-dimensional image of a target molar area is output, and the target molar comprises a first maxillary molar and / or a second maxillary molar; judging whether a second mesial buccal root canal exists in the regional three-dimensional image or not by using a deep learning method; if the second mesial cheek root canal exists, segmenting a second mesial cheek root canal image from the regional three-dimensional image by using a deep learning method; and measuring the length of the second mesial buccal root canal of the target molar by using an automatic measurement method according to the segmented second mesial buccal root canal image. According to the invention, the MB2 can be accurately identified and the length of the MB2 can be estimated, a rapid and accurate clinical diagnosis auxiliary tool is provided for a dentist, and the success rate of root canal therapy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a second mesiobuccal root canal measurement method, device and storage medium. Background Art

[0002] Maxillary molars are located at the back of the dental arch and are one of the main chewing teeth. They bear a lot of occlusal pressure. Long-term occlusal pressure can easily cause microcracks or wear of teeth, which in turn leads to damage or infection of the pulp. Due to its posterior position, oral cleaning is more difficult, and food residues are easily retained on the occlusal and adjacent surfaces of maxillary molars, leading to the occurrence of caries. After caries invade the pulp, pulpitis will occur. In addition, if apical inflammation occurs in the maxillary molar area, the inflammation may further develop into an apical abscess (pus pockets appear in the gums), which may rupture and cause the spread of infection. Severe apical inflammation may spread to the surrounding bone tissue and cause osteomyelitis of the jaw. If the affected teeth are not treated in time, the maxillary molars will be lost, and the adjacent teeth may gradually tilt or shift toward the missing area. This tilt will destroy the normal arrangement and occlusal relationship of the teeth, resulting in an unbalanced bite. The teeth and gums near the missing teeth may be subjected to abnormal pressure, increasing the risk of periodontitis and gingival atrophy. In addition, after tooth loss, the alveolar bone in the edentulous area may gradually absorb and shrink due to the lack of normal stimulation of the tooth root. This bone loss will make future restorative treatments (such as dental implants) difficult, because insufficient bone may require additional bone grafting to rebuild the alveolar bone. Therefore, maxillary molars have a higher probability of requiring root canal therapy (RCT) to improve the retention rate of teeth.

[0003] The anatomical structure of maxillary molars is complex, especially in terms of their root canal system, which usually has three main root canals: mesiobuccal root (MB), distobuccal root (DB) and palatal root (P). In addition, there is a hidden MB2 root canal (second mesiobuccal root canal, or mesiobuccal second root canal) in the maxillary first and second molars. Due to the diverse shapes and high curvature of these root canals, especially the MB2 root canal, which is usually located deeper in the mesiobuccal root, is hidden and difficult to find, and is often ignored or improperly treated in clinical treatment. These complex structures increase the risk of bacterial infection and treatment failure, which in turn leads to more root canal treatment needs.

[0004] Cone beam computed tomography (CBCT) provides three-dimensional imaging of the internal structure of the tooth with high resolution and accuracy, while reducing the superposition and deformation of the surrounding anatomical structures. CBCT can study and clinically evaluate the anatomical morphology of the root canal in a relatively short time and with a low radiation dose. However, because the MB2 root canal is usually small and close to other root canal structures, image noise, overlapping images, and the limitations of CBCT resolution may result in unclear root canal entrances or blurred root canal paths, making it difficult for doctors to accurately locate and determine the direction of the MB2 root canal. Therefore, untreated MB2 canals can lead to bacterial proliferation, causing persistent pain and even apical damage to patients. In addition, even if the MB2 is successfully found, its curved and narrow shape makes it difficult for instruments to enter and perform effective cleaning and filling, resulting in RCT failure.

[0005] Currently, most of the oral imaging analysis tools on the market focus on caries detection and periodontal disease diagnosis, and lack analysis tools for the complex anatomical structure of the root canal system, especially for MB2. Most existing image processing software still relies on manual exploration by clinicians and lacks intelligent methods to automatically identify the location and path of MB2. Therefore, the development of a more complete evaluation model is of great significance for the clinical positioning and path visualization of MB2. Summary of the invention

[0006] The present invention provides a second mesio-buccal root canal measurement method, device and storage medium to solve the technical problem that the existing image processing software relies on manual exploration by clinicians and lacks an intelligent way to automatically identify MB2.

[0007] According to one aspect of the present invention, a second mesiobuccal root canal measurement method is provided, comprising: obtaining an oral CBCT image of a detection object, obtaining a target molar position from the oral CBCT image using a deep learning method, and outputting a three-dimensional image of the target molar region, wherein the target molar includes a maxillary first molar and / or a maxillary second molar; using a deep learning method to determine whether a second mesiobuccal root canal exists in the three-dimensional image of the region; if a second mesiobuccal root canal exists, using a deep learning method to segment a second mesiobuccal root canal image from the three-dimensional image of the region; and using an automated measurement method to measure the length of the second mesiobuccal root canal of the target molar based on the segmented second mesiobuccal root canal image.

[0008] Optionally, a deep learning method is used to obtain the target molar position from the oral CBCT image, including: using an adjusted first V-NET architecture to extract three-dimensional image features in the oral CBCT image, wherein the depth of the first V-NET architecture is increased from the original 4 layers to 6 layers to accommodate a larger input image size; performing binary classification on each voxel in the oral CBCT image using a Softmax function on the last layer of feature maps, setting the voxels belonging to the target molars to one category, and setting the voxels that do not belong to the target molars to another category; and identifying the target molar position in the oral CBCT image based on the classification results.

[0009] Optionally, before using the adjusted first V-NET architecture to extract the three-dimensional image features in the oral CBCT image, the method also includes: normalizing the oral CBCT image, normalizing the size of the oral CBCT image to 512×512×512, and normalizing the voxel intensity of the oral CBCT image to the interval of [0, 1].

[0010] Optionally, a deep learning method is used to determine whether there is a second mesiobuccal root canal in the three-dimensional image of the region, including: using an adjusted three-dimensional ResNet50 architecture to perform binary classification on the three-dimensional image of the region, one category is the presence of the second mesiobuccal root canal, and the other category is the absence of the second mesiobuccal root canal, wherein the adjusted three-dimensional ResNet50 architecture refers to connecting a fully connected layer for the binary classification task after the average pooling layer.

[0011] Optionally, before using the adjusted three-dimensional ResNet50 architecture to perform binary classification on the regional three-dimensional image, the method further includes: normalizing the regional three-dimensional image to a size of 64×64×64, and using zero-padding to fill the size that does not meet the standard with 0.

[0012] Optionally, a deep learning method is used to segment a second mesiobuccal root canal image from the regional three-dimensional image, including: using an adjusted second V-NET architecture to extract the second mesiobuccal root canal edge features, wherein the depth of the second V-NET architecture is reduced from the original 4 layers to 3 layers to adapt to a smaller input image size; judging whether each voxel in the regional three-dimensional image belongs to the second mesiobuccal root canal by a Softmax function on the last layer of feature maps; segmenting the voxels belonging to the second mesiobuccal root canal in the regional three-dimensional image to obtain the second mesiobuccal root canal image.

[0013] Optionally, based on the segmented second mesiobuccal root canal image, an automated measurement method is used to measure the length of the second mesiobuccal root canal of the target molar, including: using the Zhang-Suen algorithm to iteratively refine the foreground pixels in the segmented second mesiobuccal root canal image into a skeleton line with a single pixel width; and calculating the length of the second mesiobuccal root canal based on the number of voxels and the length ratio of the skeleton line.

[0014] Optionally, calculating the length of the second mesiobuccal root canal according to the number of voxels and the length ratio of the skeleton line includes: calculating the length of the second mesiobuccal root canal using the following formula: Length=Number of Voxel×Voxel Size; Wherein, Length represents the length of the second mesiobuccal root canal of the detected object, Number of Voxel represents the number of voxels of the skeleton line, and Voxel Size represents the size of the voxel.

[0015] According to another aspect of the present invention, a second mesiobuccal root canal measurement device is also provided, comprising: a target molar detection module, used to obtain an oral CBCT image of a detection object, use a deep learning method to obtain a target molar position from the oral CBCT image, and output a three-dimensional image of a target molar region, wherein the target molar includes a maxillary first molar and / or a maxillary second molar; a second mesiobuccal root canal discrimination module, used to use a deep learning method to determine whether a second mesiobuccal root canal exists in the three-dimensional image of the region; a second mesiobuccal root canal segmentation module, used to use a deep learning method to segment a second mesiobuccal root canal image from the three-dimensional image of the region if a second mesiobuccal root canal exists; a second mesiobuccal root canal measurement module, used to use an automated measurement method to measure the length of the second mesiobuccal root canal of the target molar based on the segmented second mesiobuccal root canal image.

[0016] According to another aspect of the present invention, a storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method are implemented.

[0017] The present invention uses deep learning methods to automatically locate and measure MB2 root canals in CBCT images, which can achieve accurate recognition and analysis of complex anatomical structures. This automated technology will greatly improve the efficiency of root canal treatment and help doctors make more scientific and accurate clinical decisions through standardized measurement and recognition results. The intelligent decision support system not only improves the accuracy of diagnosis, but also provides real-time guidance and feedback in actual clinical operations, which is not available in traditional imaging technology.

[0018] In addition, in order to solve the problem that the first and second maxillary molars account for a small proportion in the image (there are a large number of invalid voxels) and many patients lack MB2 (data imbalance), the present invention only retains the molar area image in the oral CBCT image, removes useless voxels, and then determines whether there is MB2 in the area. If not, no instance segmentation of MB2 is performed, thereby ensuring that the data input to the segmentation network are all images with MB2, avoiding the problem of data imbalance. This series of processing operations can effectively improve the accuracy of segmentation and further improve the accuracy of length estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 is a flow chart of a second mesiobuccal root canal measurement method according to an embodiment of the present invention; Figure 2 A schematic diagram of a process in an implementation scenario according to an embodiment of the present invention; Figure 3 4 is a structural block diagram of a second mesiobuccal root canal measuring device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only embodiments of a part of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0022] The terms "comprises" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusions. For example, a process, method, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0023] In addition, it should be noted that the terms "first" and "second" in the "first V-NET architecture" and "second V-NET architecture" mentioned in the specification and claims of this application are identifiers used to distinguish similar objects (i.e., V-NET architectures), and do not have to be used to describe specific names, orders or sequences. The use of these terms is only to distinguish different V-NET architectures and does not contain any other additional meanings or limitations.

[0024] The "first" and "second" in the "first molar", "second molar" and "second mesiobuccal root canal" mentioned in the specification, claims and drawings of this application are specific names in medical professional terms. These terms have clear meanings and directions in the medical field and follow the norms of medical terminology. They are completely different in nature and purpose from the "first" and "second" used to distinguish the V-NET architecture mentioned above. It is hoped that such an explanation can eliminate possible misunderstandings and clarify the specific meanings and usages of the relevant terms in this application.

[0025] Example 1: Reference Figure 1 , Figure 1 is a flow chart of a second mesiobuccal root canal measurement method according to an embodiment of the present invention. Figure 1 As shown, the method includes: S1, obtaining an oral CBCT image of the test subject, obtaining a target molar position from the oral CBCT image using a deep learning method, and outputting a three-dimensional image of the target molar area, wherein the target molar includes a maxillary first molar and / or a maxillary second molar; This step is molar target detection, which aims to identify and extract the target molar image from the oral CBCT image, wherein the target molar includes the maxillary first molar and / or the maxillary second molar, that is, the target molar can be the maxillary first molar, the maxillary second molar, or a combination of the two. In some cases, the target molar includes only the maxillary first molar, or only the maxillary second molar. In some cases, the target molar can also include both the maxillary first molar and the maxillary second molar. For example, in some treatment cases, only the maxillary first molar needs root canal treatment. At this time, the target molar is the maxillary first molar. It is only necessary to identify and extract the image of the maxillary first molar, while the maxillary second molar is not processed. In some cases, if both the maxillary first molar and the maxillary second molar need to be considered or processed, the target molar includes both. The purpose of using "and / or" is to cover all the above possibilities to ensure the accuracy and comprehensiveness of the statement.

[0026] Among them, the implementation method of using deep learning methods to obtain and output the three-dimensional image of the maxillary target molar area from oral CBCT images can be: collect and label a large number of CBCT images containing target molars as training data, select a deep learning model suitable for three-dimensional image analysis, such as 3D CNN, 3D U-Net, V-NET and other models, use the labeled data to train the deep learning model so that it can accurately identify and segment the position of the target molar, and use the validation set to evaluate the performance of the model to ensure the accuracy of the model, and then use the trained deep learning model to automatically locate and extract the three-dimensional data of the target molar area from the acquired CBCT images of the detection object, and use three-dimensional visualization tools (such as VTK, Mayavi, etc.) to convert the extracted three-dimensional data into a visualized three-dimensional image.

[0027] Further, in an implementation manner of this embodiment, step S1, using a deep learning method to obtain a target molar position from the oral CBCT image, includes: S11, extracting three-dimensional image features from the oral CBCT image using the adjusted first V-NET architecture, wherein the depth of the first V-NET architecture is increased from the original 4 layers to 6 layers to accommodate a larger input image size; S12, performing binary classification on each voxel in the oral CBCT image by using a Softmax function on the last layer of feature maps, setting the voxels belonging to the target molars to one category, and setting the voxels not belonging to the target molars to another category; S13, identifying the target molar position in the oral CBCT image according to the classification result.

[0028] V-NET is a fully convolutional neural network architecture for 3D medical image segmentation. It gradually extracts and fuses image features through alternating convolution (downsampling) and upsampling (convolution transpose or upconvolution) layers, ultimately achieving accurate segmentation of 3D images.

[0029] The network design of the adjusted first V-NET architecture in this embodiment includes: maintaining the original architecture unchanged, setting the convolution kernel size to 5×5×5, with a step size of 1, setting the "lower" convolution kernel size to 2×2×2, with a step size of 2, and setting the "upper" convolution kernel size to 2×2×2, with a step size of 2. When using the convolution kernel for convolution, the image needs to be padded with a padding of 2 to ensure the same output dimension. The depth of V-NET is increased from the original 4 layers to 6 layers to adapt to the larger input image size, and the deeper network can capture more complex image features. After the last layer of up-convolution is completed, the output feature map is classified into two categories through Softmax. The Softmax function converts the score of each voxel in the feature map into a probability value, which indicates the probability that the voxel belongs to the target molar or does not belong to the target molar. The voxels belonging to the target molar are set to one category (usually marked as 1 or positive class), and the voxels not belonging to the target molar are set to another category (usually marked as 0 or negative class). Find all voxels marked as belonging to the target molar, and determine the position of the target molar according to their position and distribution in the image.

[0030] Through the above network design, the adjusted first V-Net can effectively extract features from three-dimensional medical images and achieve accurate recognition of target molars through the Softmax classifier.

[0031] Further, S11, before extracting the three-dimensional image features in the oral CBCT image using the adjusted first V-NET architecture, the method further includes: S10, normalizing the oral CBCT image, normalizing the size of the oral CBCT image to 512×512×512, and normalizing the voxel intensity of the oral CBCT image to the interval of [0, 1].

[0032] The specific steps of normalizing the size to 512×512×512 include: obtaining the original resolution of the oral CBCT image, which can be obtained by viewing the metadata of the image or using image processing software; downsampling is performed on images with an original resolution higher than 512voxel to reduce the image resolution, otherwise upsampling is performed to increase the image resolution, thereby normalizing the size of the oral CBCT image to 512×512×512.

[0033] The specific steps of normalizing the voxel intensity to the interval of [0, 1] include: determining the maximum value (Imax) and minimum value (Imin) of the voxel intensity in the oral CBCT image, and applying the following normalization formula to convert the intensity of each voxel to the interval of [0, 1]: ; where I is the original voxel intensity and Inormalized is the normalized voxel intensity.

[0034] Normalizing for size and voxel intensity helps standardize image data so that different images can be compared and analyzed at the same scale.

[0035] Through the operation of step S1 of the embodiment of the present invention, the maxillary target molar can be effectively identified from the oral CBCT image. Since the target molar only occupies a small part of the entire oral image, after identifying it, the next step of operation on the small part of the image can effectively improve the accuracy of MB2 recognition and measurement.

[0036] S2, using a deep learning method to determine whether there is a second mesiobuccal root canal in the three-dimensional image of the region; Since not every maxillary first molar or maxillary second molar has MB2, this embodiment also includes determining the presence of MB2. Step S2 is intended to determine whether MB2 exists in the three-dimensional image of the cut-out region, that is, to determine whether the target molar region obtained in step S1 has a second mesio-buccal root canal.

[0037] Among them, the implementation method of using the deep learning method to determine whether there is a second mesiobuccal root canal in the three-dimensional image of the target molar area can be: collect a large amount of three-dimensional image data containing the maxillary first and / or second molars, especially those tooth images in which the presence or absence of the second mesiobuccal root canal (MB2) has been confirmed, select a suitable deep learning model, such as a convolutional neural network (CNN), etc., use the preprocessed image data to train the model, let the model learn how to identify the presence of the second mesiobuccal root canal, apply the trained deep learning model to the three-dimensional image of the target molar obtained in step S1, and determine whether there is a second mesiobuccal root canal.

[0038] Further, in an implementation of this embodiment, S2, using a deep learning method to determine whether there is a second mesio-buccal root canal in the three-dimensional image of the region, includes: S22, using the adjusted three-dimensional ResNet50 architecture to perform two-class classification on the three-dimensional image of the region, one classifying the second mesiobuccal root canal as existing and the other classifying the second mesiobuccal root canal as not existing.

[0039] The adjusted three-dimensional ResNet50 design includes: maintaining the original architecture unchanged, the network structure and the number of parameters remain unchanged, and connecting a fully connected layer of num_channel×2 dimensions after the average pooling layer to classify the image into two categories, with MB2 as one category and without MB2 as another category, where num_channel refers to the number of channels of the feature map of the last layer.

[0040] Furthermore, before using the adjusted three-dimensional ResNet50 architecture to perform binary classification on the regional three-dimensional image, the method also includes normalizing the regional three-dimensional image to a size of 64×64×64, and using zero-padding to fill the size that does not meet the standard with 0.

[0041] Zero-padding is a commonly used padding method that uses 0 values ​​to fill the borders of an image. When the size of the original image is smaller than 64×64×64, zero-padding is used to add zero-value pixels around the image border so that the image can be expanded to the desired size. By normalizing the segmented maxillary molar area image to a size of 64×64×64, the consistency of the input data can be ensured, improving the performance and stability of the deep learning model.

[0042] Since only a small number of patients have MB2, performing instance segmentation of MB2 on all images may result in an inability to effectively identify the features of MB2. After classification in step S2 above, only those images with MB2 are segmented, which can improve detection efficiency and accuracy.

[0043] S3, if there is a second mesiobuccal root canal, segmenting the second mesiobuccal root canal image from the three-dimensional image of the region using a deep learning method; This step is MB2 image segmentation, which aims to segment the second mesio-buccal root canal image from the three-dimensional image of the target molar area where MB2 exists in step S2.

[0044] Among them, the implementation method of using a deep learning method to segment the second mesio-buccal root canal image from the three-dimensional image of the target molar area can be: collect a three-dimensional image dataset containing the maxillary first and / or second molars, and the position of MB2 should be clearly marked in these images, select a deep learning model architecture suitable for three-dimensional image segmentation, such as 3D U-Net, 3D V-Net or their variants, adjust the model parameters such as depth, width and convolution kernel size according to the size and complexity of the dataset, ensure that the output layer of the model matches the desired segmentation result, use the preprocessed dataset to train the model, and use the trained model to segment the second mesio-buccal root canal image from the three-dimensional image of the first and / or second molar area.

[0045] Further, in one implementation of this embodiment, S3, using a deep learning method to segment a second mesiobuccal root canal image from the three-dimensional image of the region, includes: S31, extracting the second mesio-buccal root canal edge feature using the adjusted second V-NET architecture, wherein the depth of the second V-NET architecture is reduced from the original 4 layers to 3 layers to adapt to a smaller input image size; S32, judging whether each voxel in the three-dimensional image of the region belongs to the second mesiobuccal root canal by using a Softmax function on the last layer of feature maps; S33, segmenting the voxels belonging to the second mesiobuccal root canal in the three-dimensional image of the region to obtain a second mesiobuccal root canal image.

[0046] The image of the molar region with MB2 is input into the adjusted V-NET, and the outline of MB2 can be identified. Among them, the design of the adjusted second V-NET in this embodiment includes: maintaining the original architecture unchanged, the size of the convolution kernel is the same as the size of the first V-NET in step S1, and reducing the depth from the original 4 layers to 3 layers to adapt to the smaller input image size. After the last layer of up-convolution is completed, the output feature map is binary classified by the Softmax function, and the voxels belonging to MB2 are set to one category, and the voxels that do not belong to MB2 are all set to another category. The identified MB2 outline is marked with different colors in the overall three-dimensional image, so that the specific position of M2 can be easily identified and MB2 can be effectively segmented.

[0047] S4, measuring the length of the second mesiobuccal root canal of the target molar using an automated measurement method according to the segmented second mesiobuccal root canal image.

[0048] This step is to measure the length of MB2.

[0049] Among them, the implementation method of measuring the length of the second mesiobuccal root canal using an automated measurement method can be: using an edge detection algorithm (such as a Canny edge detector) to identify the boundary of MB2 in the second mesiobuccal root canal image, and then using a boundary tracking algorithm or a morphological operation to extract the complete contour of MB2, and based on the extracted contour, using a distance measurement algorithm (such as Euclidean distance) to calculate the distance between two points on the contour, thereby obtaining the length of MB2.

[0050] Further, in one implementation of this embodiment, S4, measuring the length of the second mesio-buccal root canal of the target molar using an automated measurement method according to the segmented second mesio-buccal root canal image, comprises: S41, the Zhang-Suen algorithm is used to iteratively refine the foreground pixels in the segmented second mesiobuccal root canal image into skeleton lines of single pixel width; The Zhang-Suen thinning algorithm is an image thinning algorithm that can gradually thin the foreground pixels in a binary image into skeleton lines with a single pixel width. The algorithm achieves thinning by iteratively deleting pixels that meet specific conditions. Specifically, it includes: S411, initialization; The input is three-dimensional binary volume data, where the foreground voxel value is 1 and the background voxel value is 0.

[0051] Defining voxel neighbors: In three-dimensional space, each voxel has 26 neighbors (i.e., its 6 adjacent face neighbors, 12 edge neighbors, and 8 corner neighbors).

[0052] S412, iterative process; The iterative process consists of two main sub-steps, each of which gradually removes eligible voxels. Each sub-step checks and removes voxels in certain directions in turn. Sub-step 1: Iterate over all voxels and check whether each voxel satisfies the following deletion conditions: Condition 1: P is a foreground voxel (P value is 1).

[0053] Condition 2: P has at least two neighbor voxels (make sure this voxel is inside the object and not isolated).

[0054] Condition 3: The neighborhood connectivity of P remains unchanged: After deleting P, the connectivity of the objects should not be destroyed. That is, removing P does not cause the objects to be disconnected.

[0055] Condition 4: Among P’s neighbors, there is at least one background voxel in a specific direction, ensuring that deleting this voxel will not cause the skeleton to break.

[0056] For all voxels that meet the above conditions, they are marked for deletion.

[0057] Sub-step 2: For the remaining voxels, we iterate again and apply a similar set of conditions, but in this substep we check different directional neighbors of the voxel.

[0058] Delete the voxels that meet the deletion criteria again.

[0059] S413, stop iteration; The iterative process ends when no further voxels can be deleted in all directions. In other words, the algorithm ends when all eligible voxels have been removed.

[0060] S42, calculating the length of the second mesiobuccal root canal according to the number of voxels and the length ratio of the skeleton line; Among them, according to the number of voxels and the length ratio of the skeleton line, the implementation method for calculating the length of the second mesiobuccal root canal can be: traverse the obtained skeleton line, record the position of each voxel, for each voxel on the skeleton line, calculate the distance between it and the adjacent voxel, accumulate the distances between all adjacent voxels to obtain the total length of the skeleton line, and then apply the length ratio factor to the total length of the entire skeleton line to obtain the length of the second mesiobuccal root canal.

[0061] Further, in an implementation of this embodiment, after the skeleton line is obtained, the MB2 length is estimated according to the number of voxels occupied by the skeleton line, wherein the calculation formula is: Length=Number of Voxel×Voxel Size; Wherein, Length represents the length of the second mesiobuccal root canal of the detected object, Number of Voxel represents the number of voxels of the skeleton line, and Voxel Size represents the size of the voxel, which is converted according to the CBCT image data.

[0062] In view of the problem that the first and second maxillary molars account for a small proportion in the image (there are a large number of invalid voxels) and many patients lack MB2 (data imbalance), the embodiment of the present invention proposes a solution for the measurement of the second mesiobuccal root canal, aiming to develop a model for the automatic positioning and measurement of the MB2 (second mesiobuccal root canal) of the first and / or second maxillary molars by combining deep learning technology with medical image processing technology. The method is divided into a molar target detection part, a judgment part on whether MB2 exists, a segmentation part of MB2, and a part of estimating the length of MB2. In the molar target detection part, the normalized CBCT data is input into the adjusted V-NET to detect the position of the tooth. First, the maxillary first and / or second molar area is divided, and only the first and / or second molars (i.e., target molars) and the images near them are retained, and useless voxels are eliminated. In the judgment of whether MB2 exists, the standardized tooth area image is input into the adjusted three-dimensional ResNet50 to perform binary classification on the molar area image to determine whether the retained image area has MB2. If not, the instance segmentation of MB2 is not performed, thereby ensuring that the data input into the segmentation network are all images with MB2, avoiding data imbalance problems. This series of preprocessing operations can effectively improve the accuracy of segmentation and further improve the accuracy of length estimation. In the segmentation of MB2, the standardized tooth area image is input into the adjusted V-NET to segment the contour of MB2. In the estimation of MB2 length, the segmented contour image is skeletonized into skeleton lines and the length is estimated by the number of voxels it occupies.

[0063] The embodiment of the present invention can accurately identify MB2 and estimate its length, providing dentists with a fast and accurate clinical diagnosis auxiliary tool, thereby improving the success rate of root canal treatment.

[0064] Embodiment 2: The technical solution of the present invention is further described in detail below in conjunction with a specific embodiment and the accompanying drawings. It should be understood that the following embodiment is only used to explain the present invention and is not used to limit the present invention.

[0065] Reference Figure 2 , Figure 2 Schematic diagram of a process in an implementation scenario according to an embodiment of the present invention. Figure 2 As shown, the method includes: The original images of oral CBCT image data were normalized, and the normalized image size was 512x512x512; The normalized oral CBCT images were input into the adjusted V-NET network to obtain the images of the maxillary first and second molar areas; The images of the maxillary first and second molar regions were normalized again, and the normalized size was 64x64x64 images; The normalized images of the maxillary first and second molar regions are input into the adjusted three-dimensional ResNet50 model to determine whether there is MB2 in the molar region (in some embodiments, this step can be skipped directly, and the normalized images of the maxillary first and second molar regions are directly input into the adjusted V-NET network to segment the MB2 image). If it is determined that there is no MB2, the process ends; If it is determined that there is MB2, the images of the maxillary first and second molar regions with MB2 are input into the adjusted V-NET network to segment the MB2 image; The MB2 segmented image is input into the 3D Zhang-Suen model to extract the skeleton image; The MB2 length is calculated using the MB2 length calculation model.

[0066] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0067] Embodiment 3: In this embodiment, a second mesio-buccal root canal measuring device is also provided, which is used to implement the above embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the terms "module" and "unit" can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0068] Figure 3 is a structural block diagram of a second mesiobuccal root canal measuring device according to an embodiment of the present invention. Figure 3 As shown, the device includes: a target molar detection module 100, a second mesiobuccal root canal discrimination module 200, a second mesiobuccal root canal segmentation module 300, and a second mesiobuccal root canal measurement module 400, wherein: The target molar detection module 100 is used to obtain an oral CBCT image of the detection object, obtain the target molar position from the oral CBCT image using a deep learning method, and output a three-dimensional image of the target molar area, wherein the target molar includes the maxillary first molar and / or the maxillary second molar; A second mesiobuccal root canal identification module 200 is used to determine whether there is a second mesiobuccal root canal in the three-dimensional image of the region by using a deep learning method; A second mesiobuccal root canal segmentation module 300 is used to segment a second mesiobuccal root canal image from the three-dimensional image of the region using a deep learning method if the second mesiobuccal root canal exists; The second mesiobuccal root canal measurement module 400 is used to measure the length of the second mesiobuccal root canal of the target molar using an automated measurement method according to the segmented second mesiobuccal root canal image.

[0069] Optionally, the target molar detection module includes: an image feature extraction unit, used to extract three-dimensional image features in the oral CBCT image using an adjusted first V-NET architecture, wherein the depth of the first V-NET architecture is increased from the original 4 layers to 6 layers to accommodate a larger input image size; a target molar identification unit, used to perform binary classification on each voxel in the oral CBCT image through a Softmax function on the last layer of feature maps, setting the voxels belonging to the target molars to one category and the voxels not belonging to the target molars to another category; and identifying the target molar position in the oral CBCT image according to the classification results.

[0070] Optionally, the target molar detection module also includes: a CBCT image normalization processing unit, used to normalize the oral CBCT image, normalize the size of the oral CBCT image to 512×512×512, and normalize the voxel intensity of the oral CBCT image to the interval of [0, 1].

[0071] Optionally, the second mesiobuccal root canal discrimination module includes: a second mesiobuccal root canal identification unit, used to use an adjusted three-dimensional ResNet50 architecture to perform two-class classification on the three-dimensional image of the region, one class is the presence of a second mesiobuccal root canal, and the other class is the absence of a second mesiobuccal root canal, wherein the adjusted three-dimensional ResNet50 architecture refers to a fully connected layer for a two-classification task connected after the average pooling layer.

[0072] Optionally, the second mesio-buccal root canal identification module further includes: a regional three-dimensional image normalization processing unit, used to normalize the regional three-dimensional image, normalize the regional three-dimensional image to a size of 64×64×64, and use zero-padding to fill the size that does not meet the standard with 0.

[0073] Optionally, the second mesiobuccal root canal segmentation module includes: an edge feature extraction unit, used to extract the second mesiobuccal root canal edge feature using the adjusted second V-NET architecture, wherein the depth of the second V-NET architecture is reduced from the original 4 layers to 3 layers to adapt to a smaller input image size; a second mesiobuccal root canal segmentation unit, used to determine whether each voxel in the three-dimensional image of the region belongs to the second mesiobuccal root canal through a Softmax function on the last layer of feature map; and segment the voxels belonging to the second mesiobuccal root canal in the three-dimensional image of the region to obtain a second mesiobuccal root canal image.

[0074] Optionally, the second mesiobuccal root canal measurement module includes: a skeleton line construction unit, used to iteratively refine the foreground pixels in the segmented second mesiobuccal root canal image into a skeleton line of single pixel width using the Zhang-Suen algorithm; a second mesiobuccal root canal measurement unit, used to calculate the length of the second mesiobuccal root canal based on the number of voxels and length ratio of the skeleton line.

[0075] Optionally, the second mesiobuccal root canal measurement unit is used to calculate the length of the second mesiobuccal root canal using the following formula: Length=Number of Voxel×Voxel Size; wherein Length represents the length of the second mesiobuccal root canal of the detection object, Number of Voxel represents the number of voxels of the skeleton line, and Voxel Size represents the size of the voxel.

[0076] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0077] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store computer programs.

[0078] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0079] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.

[0080] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0081] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0082] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0083] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0084] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0085] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc., which can store program code.

[0086] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for measuring the second mesio-buccal root canal, characterized in that: include: Acquire an oral CBCT image of the test subject, use a deep learning method to acquire a target molar position from the oral CBCT image, and output a three-dimensional image of the target molar region, wherein the target molar includes a maxillary first molar and / or a maxillary second molar; Using a deep learning method to determine whether there is a second mesiobuccal root canal in the three-dimensional image of the region; If a second mesiobuccal root canal exists, segmenting the second mesiobuccal root canal image from the three-dimensional image of the region using a deep learning method; According to the segmented second mesiobuccal root canal image, an automated measurement method is used to measure the length of the second mesiobuccal root canal of the target molar.

2. The second mesiobuccal root canal measurement method according to claim 1, characterized in that: The target molar position is obtained from the oral CBCT image using a deep learning method, including: Extracting three-dimensional image features from the oral CBCT image using the adjusted first V-NET architecture, wherein the depth of the first V-NET architecture is increased from 4 layers to 6 layers to accommodate a larger input image size; On the last layer of feature maps, each voxel in the oral CBCT image is classified into two categories by using a Softmax function, where the voxels belonging to the target molars are set to one category, and the voxels not belonging to the target molars are set to another category; The target molar position in the oral CBCT image is identified according to the classification result.

3. The second mesiobuccal root canal measurement method according to claim 2, characterized in that: Before extracting the three-dimensional image features in the oral CBCT image using the adjusted first V-NET architecture, the method further includes: The oral CBCT image is normalized to normalize the size of the oral CBCT image to 512×512×512, and the voxel intensity of the oral CBCT image is normalized to the interval of [0, 1].

4. The second mesiobuccal root canal measurement method according to claim 1, characterized in that: Using a deep learning method to determine whether there is a second mesiobuccal root canal in the three-dimensional image of the region includes: The three-dimensional image of the region is classified into two categories using the adjusted three-dimensional ResNet50 architecture, one category is the presence of the second mesio-buccal root canal, and the other category is the absence of the second mesio-buccal root canal, wherein the adjusted three-dimensional ResNet50 architecture refers to connecting a fully connected layer for the two-category task after the average pooling layer.

5. The second mesiobuccal root canal measurement method according to claim 4, characterized in that: Before using the adjusted three-dimensional ResNet50 architecture to perform binary classification on the regional three-dimensional image, the method further includes: The three-dimensional image of the region is normalized to a size of 64×64×64, and a size that does not meet the standard is filled with zero using zero-padding.

6. The second mesiobuccal root canal measurement method according to claim 1, characterized in that: Segmenting a second mesio-buccal root canal image from the three-dimensional image of the region using a deep learning method, including: The second mesio-buccal root canal edge features were extracted using the adjusted second V-NET architecture, where the depth of the second V-NET architecture was reduced from 4 layers to 3 layers to adapt to a smaller input image size; On the last layer of feature maps, a Softmax function is used to determine whether each voxel in the three-dimensional image of the region belongs to the second mesiobuccal root canal; The voxels belonging to the second mesiobuccal root canal in the three-dimensional image of the region are segmented to obtain a second mesiobuccal root canal image.

7. The second mesiobuccal root canal measurement method according to claim 1, characterized in that: According to the segmented second mesiobuccal root canal image, an automated measurement method is used to measure the length of the second mesiobuccal root canal of the target molar, including: The Zhang-Suen algorithm was used to iteratively refine the foreground pixels in the segmented second mesiobuccal root canal image into skeleton lines with a single pixel width. The length of the second mesiobuccal root canal was calculated based on the number of voxels and length ratio of the skeleton line.

8. The second mesiobuccal root canal measurement method according to claim 7, characterized in that: According to the number of voxels and length ratio of the skeleton line, the length of the second mesiobuccal root canal is calculated, including: The length of the second mesiobuccal canal was calculated using the following formula: Length=Number of Voxel×Voxel Size; Wherein, Length represents the length of the second mesiobuccal root canal of the detected object, Number of Voxel represents the number of voxels of the skeleton line, and Voxel Size represents the size of the voxel.

9. A second mesio-buccal root canal measuring device, characterized in that: include: A target molar detection module is used to obtain an oral CBCT image of the detection object, obtain the target molar position from the oral CBCT image using a deep learning method, and output a three-dimensional image of the target molar area, wherein the target molar includes the maxillary first molar and / or the maxillary second molar; A second mesiobuccal root canal discrimination module, used for judging whether there is a second mesiobuccal root canal in the three-dimensional image of the region by using a deep learning method; a second mesiobuccal root canal segmentation module, configured to segment a second mesiobuccal root canal image from the three-dimensional image of the region using a deep learning method if the second mesiobuccal root canal exists; The second mesiobuccal root canal measurement module is used to measure the length of the second mesiobuccal root canal of the target molar using an automated measurement method according to the segmented second mesiobuccal root canal image.

10. A storage medium, characterized in that: The storage medium stores a computer program, which implements the steps of the method according to any one of claims 1 to 8 when executed by a processor.

Images