Method and device for detecting maturity of zygomatic maxillary seam and medium

Abstract

The invention relates to a method and device for detecting the maturity of a zygomatic maxillary seam and a medium, and the method comprises the steps: providing a first three-dimensional image corresponding to the zygomatic maxillary part of a to-be-examined person, and enabling the first three-dimensional image to be obtained based on the medical imaging of the zygomatic maxillary part of the to-be-examined person; preprocessing the first three-dimensional image to obtain a second three-dimensional image; according to the second three-dimensional image and a trained bone seam position prediction neural network, a predicted bone seam position is obtained, and the predicted bone seam position is located on voxels of the second three-dimensional image; cutting the second three-dimensional image according to the predicted bone seam position to obtain a target bone seam graph; and at least inputting the target bone seam graph into a trained bone seam maturity neural network to obtain the maturity of the bone seam.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular provides a method, a device and a medium for detecting the maturity of a zygomatico-maxillary suture. Background Art

[0002] The existing method for assessing the maturity of the zygomaticomaxillary suture is based on oral CT three-dimensional images composed of hundreds of CT images. Doctors need to manually locate the suture position and undergo complex processing before they can assess the maturity of the zygomaticomaxillary suture. Since the method for assessing the zygomaticomaxillary suture is complex, it requires doctors to have high experience and knowledge.

[0003] Accordingly, the art needs a new maturity detection scheme for the zygomaticomaxillary suture to solve the above problems. Summary of the invention

[0004] In order to overcome the above defects, the present invention is proposed to solve or at least partially solve the technical problem in the prior art that the maturity of the zygomatico-maxillary suture is difficult to judge.

[0005] In a first aspect, the present invention provides a method for detecting the maturity of a zygomatico-maxillary suture, the method comprising: providing a first three-dimensional image corresponding to the zygomatico-maxillary part of a person to be examined, wherein the first three-dimensional image is obtained based on medical imaging of the zygomatico-maxillary part of the person to be examined; preprocessing the first three-dimensional image to obtain a second three-dimensional image; obtaining a predicted suture position based on the second three-dimensional image and a trained suture position prediction neural network; cropping the second three-dimensional image according to the predicted suture position to obtain a target suture image; at least inputting the target suture image into a trained suture maturity neural network to obtain the maturity of the suture.

[0006] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, wherein the second three-dimensional image is composed of multiple prediction units, and the prediction unit includes at least one voxel, including: obtaining a predicted suture position based on the second three-dimensional image and a trained suture position prediction neural network, including: inputting the second three-dimensional image into the trained suture position prediction neural network to obtain the reliability of the prediction unit, wherein the reliability is used to indicate the degree to which the prediction unit is a suture; and obtaining a predicted suture position based on the reliability of the prediction unit.

[0007] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, preprocessing the first three-dimensional image to obtain a second three-dimensional image includes: adjusting the first three-dimensional image using preset display attributes to obtain at least a first channel image and a second channel image, wherein the display attributes corresponding to the first channel image and the second channel image are different; and superimposing at least the first channel image and the second channel image to obtain a second three-dimensional image with dual channels.

[0008] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the display attributes include window width and window level, wherein the window width of the first channel image is greater than the window width of the second channel image, and the window level of the first channel image is less than the window level of the second channel image.

[0009] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the predicted suture position is obtained based on the reliability of the prediction unit, including: sorting the prediction units according to the reliability; filtering prediction units that meet preset filtering conditions from the prediction units to obtain a filtered prediction unit set, wherein the preset filtering conditions are used to select prediction units with high reliability from the prediction units; and obtaining the predicted suture position based on the prediction unit set.

[0010] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, obtaining the predicted suture position based on the prediction unit set includes: obtaining a coordinate information set of the prediction unit in the corresponding prediction unit set; determining the predicted suture position based on the coordinate information set, wherein the coordinates corresponding to the predicted suture position include at least the median coordinates determined by the coordinate information set.

[0011] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, before the at least target suture image is input into a trained suture maturity neural network to obtain the maturity of the suture, the method also includes: inputting the target suture image into a trained rotation neural network to obtain the normal vector of the plane where the suture is located; rotating the target suture image at least based on the normal vector to obtain a rotated suture image, wherein the normal vector of the rotated suture image reaches a preset observation angle with the observation plane; the at least target suture image is input into a trained suture maturity neural network to obtain the maturity of the suture, including: inputting the target suture image and the rotated suture image into the trained suture maturity neural network to obtain the maturity of the suture.

[0012] In a second aspect, a control device is provided, which includes a processor and a storage device, wherein the storage device is suitable for storing a plurality of computer programs, and the computer programs are suitable for being loaded and run by the processor to execute the maturity detection method of the zygomaticomaxillary suture described in any one of the technical solutions of the above-mentioned technical solutions of the maturity detection method of the zygomaticomaxillary suture.

[0013] In a third aspect, a computer-readable storage medium is provided, wherein a plurality of computer programs are stored therein, wherein the computer programs are suitable for being loaded and run by a processor to execute the maturity detection method of the zygomaticomaxillary suture as described in any one of the technical solutions of the above-mentioned maturity detection method of the zygomaticomaxillary suture.

[0014] The above one or more technical solutions of the present invention have at least one or more of the following Beneficial effects:

[0015] In the technical solution of the present invention, the suture position prediction neural network is used to automatically locate the suture, and the three-dimensional image is preprocessed and intelligently cropped to ensure that the maturity assessment focuses on the most relevant image area, thereby improving the accuracy of the assessment. Finally, the suture maturity is automatically assessed by the suture maturity neural network, which enhances the consistency and reliability of the assessment results and makes the entire assessment process more efficient and standardized. In addition, the present technology improves the efficiency and accuracy of the assessment by using a step-by-step processing method and using corresponding multiple neural networks to process each process separately. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In addition, similar numbers in the figures are used to represent similar components, among which:

[0017] Figure 1 1 is a schematic flow chart of main steps of a method for detecting the maturity of the zygomatico-maxillary suture according to an embodiment of the present invention;

[0018] Figure 2 is a schematic diagram of a process of obtaining a second three-dimensional image according to a method for detecting the maturity of the zygomatico-maxillary suture according to an embodiment of the present invention;

[0019] Figure 3 It is a schematic diagram of a process of obtaining a rotational suture map according to a method for detecting the maturity of the zygomaticomaxillary suture according to an embodiment of the present invention;

[0020] Figure 4 It is another schematic flow chart of obtaining a rotated suture map of the method for detecting the maturity of the zygomaticomaxillary suture according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0022] In the description of the present invention, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as computer programs, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Non-temporary computer-readable storage media include any suitable medium that can store computer programs, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one A or B" or "at least one of A and B" has a similar meaning to "A and / or B", and may include only A, only B, or A and B. The singular terms "one" and "the" may also include plural forms.

[0023] Embodiment 1:

[0024] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart showing the main steps of a method for detecting the maturity of the zygomaticomaxillary suture according to an embodiment of the present invention. Figure 1 As shown, the maturity detection method of the zygomaticomaxillary suture in the embodiment of the present invention mainly includes the following steps S10 to S50.

[0025] Step S10: providing a first three-dimensional image corresponding to the zygomatico-maxillary region of the subject to be examined.

[0026] In this embodiment, the first three-dimensional image is obtained based on medical imaging of the zygomatic and maxillary region of the subject to be examined.

[0027] In one embodiment, the medical imaging technology used to obtain the first three-dimensional image may include but is not limited to one or more of the following:

[0028] Computed tomography (CT), which uses X-rays to scan at multiple angles to create high-resolution, three-dimensional images, is suitable for observing bone structure; magnetic resonance imaging (MRI), which uses magnetic fields and radio waves to obtain detailed images of soft tissues;

[0029] Cone beam CT (CBCT), which is specifically used for oral and craniomaxillofacial imaging, providing three-dimensional images with higher resolution and lower radiation dose;

[0030] Ultrasound imaging, which generates images through the reflection of sound waves, is suitable for the observation of soft tissues. The combination of the above medical imaging technologies ensures accurate imaging of the zygomatic and maxillary areas, providing comprehensive data support for subsequent analysis.

[0031] In this embodiment, CBCT images are used as an example for explanation. The first three-dimensional image is obtained based on multiple CBCT images. Multiple CBCT images constitute a comprehensive scan of the zygomatico-maxillary region of the subject to be examined. The CBCT image includes multiple scanned images, which together constitute the first three-dimensional image of the zygomatico-maxillary region of the subject to be examined after processing. Through this comprehensive scanning method, the detailed structure of the zygomatico-maxillary region of the subject to be examined can be upgraded from two-dimensional information to three-dimensional information, thereby providing highly comprehensive view information for subsequent analysis.

[0032] In this embodiment, the first three-dimensional image can be observed through multiple single-direction two-dimensional images, and the common single directions are one of the horizontal, dorsal, and longitudinal directions. The horizontal two-dimensional image can provide the upper and lower relationship and structural hierarchy of the zygomatic bone; the dorsal two-dimensional image shows the relative position of the zygomatic bone and other parts of the skull from back to front, which is helpful to observe the connection between the zygomatic bone and the adjacent bone structure; the lateral two-dimensional image refers to an image that captures the zygomatic-maxillary area from one side to the other (the left and right sides of the human body). Each single direction can provide cross-sectional information about one dimension of the zygomatic-maxillary structure, and these multi-dimensional information are helpful to evaluate the development stage of the zygomatic-maxillary structure.

[0033] In this embodiment, by acquiring the first three-dimensional image of the zygomaticomaxillary region of the subject to be examined, the three-dimensional information of the zygomaticomaxillary region of the subject to be examined can be accurately captured, which provides a basis for ensuring the accuracy and reliability of subsequent image preprocessing, suture positioning and maturity judgment. In addition, this detailed three-dimensional data acquisition method significantly improves the comprehensiveness and scientificity of the entire diagnostic process, ensures the accurate assessment of the maturity detection of the zygomaticomaxillary suture, and thus assists medical professionals in more accurate treatment planning and health management.

[0034] Step S20: preprocessing the first three-dimensional image to obtain a second three-dimensional image.

[0035] In this embodiment, the second three-dimensional image is composed of a plurality of prediction units, and the prediction unit includes at least one voxel. The second three-dimensional image is composed of voxels, which are different from pixels. Pixels only have two-dimensional coordinate information, while voxels have three-dimensional coordinate information.

[0036] In this embodiment, the preprocessing of the first three-dimensional image is to optimize the information content of the original scanned image through a specific image processing technique, thereby generating a second three-dimensional image. The generated second three-dimensional image makes subsequent analysis more accurate.

[0037] In one embodiment, the preprocessing process includes: one of image denoising, contrast enhancement, scaling or cropping. Image denoising is performed by using a denoising algorithm (such as Gaussian blur or median filtering) to reduce random noise in the first three-dimensional image and improve the quality of the first three-dimensional image. Contrast enhancement involves adjusting the grayscale range of the first three-dimensional image to make important features (such as sutures and other structures) in the first three-dimensional image more obvious.

[0038] In this embodiment, the prediction unit refers to the smallest unit in the second three-dimensional image or other images for analysis and prediction, and the prediction unit is also an image block. In one embodiment, the prediction unit can be a cube-shaped voxel frame composed of eight voxels, or a 2*2 cuboid voxel frame composed of four voxels with one side being a square and the other side being a rectangle. Of course, it can also be other voxel frames composed of other numbers of voxels that can densely tile the second three-dimensional image.

[0039] In one embodiment, the prediction unit is a voxel. In this embodiment, each voxel can be used independently to predict the presence or state of a suture. This high-resolution processing method enables the neural network model to analyze images at a very fine level, thereby improving the accuracy of locating sutures and assessing their maturity.

[0040] In this embodiment, by extracting as much useful information as possible from the first three-dimensional image, the information in the first three-dimensional image is enhanced, thereby providing high-quality input data for subsequent suture position prediction and maturity analysis, thereby improving the accuracy and efficiency of the entire analysis process.

[0041] In one embodiment, the second three-dimensional image is obtained through steps S201-S202, such as Figure 2 As shown, the details are as follows:

[0042] Step S201: using preset display properties to adjust a first three-dimensional image to obtain at least a first channel image and a second channel image.

[0043] In this embodiment, the first channel image and the second channel image have different display attributes.

[0044] In one embodiment, the display attributes include window width and window level, wherein the window width of the first channel image is greater than the window width of the second channel image, and the window level of the first channel image is less than the window level of the second channel image.

[0045] In this embodiment, different window width and window level parameters are used to optimize image visualization, thereby adapting to different diagnostic requirements.

[0046] In this embodiment, the window width defines the grayscale range that can be visualized in the image. A wider window width can display a wider range of grayscales, allowing the image to display more details, but may reduce contrast. On the contrary, a narrower window width provides a higher contrast, making the details of a specific grayscale more prominent, but may lose some detail information. The window position defines the central grayscale value of the window width. Adjusting the window position can move the center of the grayscale window, thereby affecting the overall brightness of the image.

[0047] In this embodiment, the first channel image is suitable for observing high-density structures such as bones due to its wide window width and low window level. The first channel image can provide more visual information about bones and hard tissues because its wider window width captures a wider range of density differences. The second channel image is more suitable for observing soft tissues due to its narrow window width and high window level. The narrow window width provides a higher contrast, which helps to distinguish soft tissue structures with similar density.

[0048] Through the above steps, the purpose of extracting maximum information from the first three-dimensional image is achieved, and the adjustment of the window width and the window level is effectively used to provide information at different angles.

[0049] Step S202: superimposing at least the first channel image and the second channel image to obtain a second three-dimensional image with two channels.

[0050] In one embodiment, voxel synthesis, image registration and normalization techniques are used to ensure correct superposition of the two images and maximize information utilization.

[0051] In one implementation, it is assumed that the window width of the first channel map is 2000 and the window level is 300. The window width of the second channel map is 1000 and the window level is 500.

[0052] For the same voxel point in the two channel images, assume that the CT value of the voxel point is 700HU. Then the grayscale value calculated by the first channel number is approximately 178.5, and the grayscale value calculated by the second channel number is approximately 114.7. When the second three-dimensional image is formed later, the grayscale value corresponding to the voxel point is (178.5, 114.7). The two channels in the second three-dimensional image correspond to the grayscale values ​​in the first channel image and the grayscale values ​​in the second channel image, respectively. If the display grayscale value of the second three-dimensional image can be calculated based on these two values, for example, averaged after normalization, and mapped in the grayscale value display range of the second three-dimensional image to obtain the grayscale image of the second three-dimensional image.

[0053] Specifically, in this embodiment, the first channel image and the second channel image are fused so as to retain the unique information of each channel during the superposition process. For example, if the first channel image is superior in displaying the details of the bone structure, while the second channel image is better in depicting the soft tissue or low-contrast area, the superposition algorithm will intelligently adjust the respective contribution weights so that the composite image has both a clear display of the bone structure and the details of the soft tissue.

[0054] In addition, in order to facilitate the subsequent processing of the neural network, normalization processing is used here when superimposing to ensure that the first channel image and the second channel image will not cause image quality variation after synthesis, thereby ensuring the quality and consistency of the synthesized image.

[0055] Step S30: obtaining a predicted suture position according to the second three-dimensional image and the trained suture position prediction neural network.

[0056] In this embodiment, the predicted suture position is located on the prediction unit of the second three-dimensional image.

[0057] In one embodiment, the predicted suture position is obtained through step S301 and step S302, as follows:

[0058] Step S301: input the second three-dimensional image into the trained suture position prediction neural network to obtain the reliability of the prediction unit.

[0059] In this embodiment, the reliability is used to indicate the degree to which the prediction unit is a suture.

[0060] In one embodiment, the suture position prediction neural network is specifically designed to predict suture positions. In this embodiment, the suture position prediction neural network outputs the reliability of the prediction unit, which quantifies the degree to which each prediction unit is likely to be a suture.

[0061] In one embodiment, the confidence is a probability value between 0 and 1, wherein a value close to 1 indicates a high confidence, i.e., the predicted unit is very likely to be part of a suture, and a value close to 0 indicates a low confidence, i.e., the predicted unit is unlikely to be part of a suture. In this embodiment, the prediction unit is defined as a single voxel.

[0062] In one embodiment, the suture position prediction neural network is a 3D U-net neural network. The training of the suture position prediction neural network uses data enhancement, such as image translation, image rotation, image scaling, gamma transform, Gaussian blur, elastic deformation and other operations. Specifically, image translation simulates the scenes of sutures in different positions by randomly moving images in three-dimensional space, which helps the network learn to identify sutures from any position. Image rotation imitates different viewing angles by rotating at random angles, so that the network can more robustly identify sutures in various directions. Image scaling simulates changes in distance by changing the size of the image to ensure that the neural network is insensitive to changes in the size of the suture. Gamma transform adjusts the brightness and contrast of the image, which helps the network maintain robustness under different lighting and contrast conditions. Gaussian blur simulates noise or focusing problems in the imaging process by introducing a blurring effect. Elastic deformation increases the diversity of samples by applying random nonlinear deformations to simulate the natural variation of biological tissues.

[0063] In addition, during the training of the suture position prediction neural network, Adaptive Wing Loss is selected as the loss function in the training process. The Adaptive Wing Loss function is more sensitive to label imbalance and error-prone prediction boundaries, and can better optimize the performance of the neural network in difficult-to-distinguish areas. Since Adaptive Wing Loss can adapt to the sensitivity of different output values, it is particularly suitable for processing small but important features in images, such as small sutures.

[0064] Through the above steps, the suture position prediction neural network can predict the reliability of the prediction unit, thus laying the foundation for the subsequent suture position. In the training process of the suture position prediction neural network, the characteristics of the suture are gradually learned by building more abstract feature representations layer by layer. Due to the particularity of the 3DU-net architecture, it retains the high-level feature information captured in the deep layer of the network through special upsampling and skip connection strategies. Combined with the choice of loss function, the trained suture position prediction neural network is more sensitive to the suture position and the reliability of the recognition is more accurate.

[0065] Step S302: Obtain the predicted suture position according to the reliability of the prediction unit.

[0066] In one implementation, the credibility can be screened by a threshold method or a TOP-N method.

[0067] In one embodiment, the reliability value is thresholded. A suitable reliability threshold can be determined by performance optimization on a training data set. Prediction units with reliability values ​​higher than this threshold are considered to be candidate regions that may belong to sutures and are retained. Next, spatial clustering analysis is performed on these candidate regions to identify continuous or adjacent sets of high-reliability prediction units. Through cluster analysis, scattered high-reliability points can be combined into continuous suture regions, which helps to further improve the accuracy and reliability of suture detection. Smoothing and morphological operations are then performed to optimize the representation of suture boundaries and eliminate possible noise points, making the final suture position more accurate.

[0068] In another embodiment, the predicted suture position is obtained through steps S302-1 to S302-3, as follows:

[0069] Step S302-1: sort the prediction units according to their reliability.

[0070] In one implementation, each prediction unit is sorted in descending order according to its confidence value, so that the prediction unit with the highest confidence is at the top of the list.

[0071] In this embodiment, the sorting process is implemented using a sorting algorithm, such as quick sort or merge sort.

[0072] Step S302 - 2 : Filter prediction units that meet preset filtering conditions from the prediction units to obtain a filtered prediction unit set.

[0073] In this embodiment, the preset screening condition is used to select a prediction unit with high reliability from the prediction units.

[0074] In one embodiment, the preset screening condition may be to set one or more confidence thresholds, such as selecting prediction units with a confidence higher than 0.8. In addition, the screening condition may also be prediction units with a confidence in the top N%, such as 5%. In this way, high confidence prediction units can be screened out.

[0075] In this embodiment, the prediction units with high confidence are selected as the basis for further analysis because these prediction units are most likely to represent the actual suture locations. By focusing on these high probability areas, the efficiency and accuracy of the entire suture detection process can be significantly improved.

[0076] This step is a decisive filtering link in the entire suture position prediction process. It selects the most representative and diagnostically valuable prediction units from a large number of candidates through conditional settings. Since this application works in stages, the processing efficiency of subsequent steps and the accuracy of the results are improved.

[0077] Step S302-3: Obtain the predicted suture position according to the prediction unit set.

[0078] In one embodiment, the selected high-confidence prediction unit set is subjected to spatial and statistical analysis to identify the most likely suture location.

[0079] In this embodiment, the geometric center is determined by the geometric center of the corresponding coordinates of the prediction unit set, which represents the most likely position of each potential suture region. The calculation of the geometric center can be simply completed by calculating the arithmetic mean of the coordinates of all prediction units in the prediction unit set.

[0080] This step accurately identifies the suture location by combining spatial distribution and prediction unit reliability.

[0081] Furthermore, in one embodiment, through steps S302-31 to S302-32, the predicted suture position is obtained, as follows:

[0082] Step S302 - 31 : Obtain a set of coordinate information of prediction units in a corresponding prediction unit set.

[0083] In one implementation, coordinate information of the prediction units in the measurement unit set is obtained, and this information is collected to form a coordinate information set.

[0084] Step S302-32: Determine the predicted suture position based on the coordinate information set.

[0085] In this embodiment, the coordinates corresponding to the predicted suture positions at least include the median coordinates determined by the coordinate information set.

[0086] In one embodiment, median coordinates are used as a method of determining suture locations.

[0087] First, the median coordinate is the coordinate that is in the middle of all the selected prediction unit coordinates after the coordinate values ​​are sorted. The process of calculating the median coordinate includes sorting all X coordinate values, all Y coordinate values, and all Z coordinate values ​​separately, and then selecting the middle value in each dimension as the final predicted position. The main advantage of this method is that it is very resistant to extreme values, so it can still provide reliable position predictions when there are outliers in the prediction unit.

[0088] The reason for choosing to use the median coordinates is that in the prediction of suture positions. Since sutures are usually long and thin structures that may pass through multiple voxels, the median method can effectively reduce the impact of accidental errors on the final predicted position, thereby improving the accuracy and reliability of diagnosis. By using the median coordinates, this step reduces the impact of outliers on the prediction of suture positions, thereby achieving the effect of improving the overall prediction accuracy.

[0089] This step determines the predicted suture position by comprehensively applying the median coordinates, which not only improves the accuracy of the prediction, but also enhances the robustness of the method, making it more reliable in application.

[0090] Step S40: cropping the second three-dimensional image according to the predicted suture positions to obtain a target suture image.

[0091] In this embodiment, the target suture image is cut out based on a plurality of second three-dimensional images. The target suture image can be a three-dimensional image, and can also be a collection of two-dimensional images in at least one single direction obtained from the three-dimensional image.

[0092] In one embodiment, once the predicted suture location is determined, the next task is to crop the second three-dimensional image so as to extract the specific area containing the suture from the second three-dimensional image, forming an image focused on the suture, and this result image is the "target suture map". In this embodiment, the target suture map is obtained based on the original image (the first three-dimensional image) and has the characteristics of the original image.

[0093] In one embodiment, during the cropping process, the size and shape of the cropping area are first determined. The cropped area is usually a three-dimensional area around the center of the predicted suture position for each second three-dimensional image, such as a cube. The size of the area must ensure that it contains the complete suture and a certain range around it, so as to facilitate subsequent detailed analysis. In one embodiment, the cropped area is a three-dimensional cuboid or a cube. The specific cropping strategy is: construct a cube with the predicted suture position as the center of the cube, retain the image inside the cube, and crop the image outside the cube. The specific size and shape of the cropping are preset based on experience.

[0094] In one embodiment, the cropping operation can be performed using image processing software or an automated script. In one embodiment, the cropping process can also integrate an error checking mechanism to prevent the situation where the suture is not completely included in the cropped image due to prediction errors.

[0095] In this embodiment, by cropping, the subsequent neural network analysis focuses only on the image area containing the suture, which can reduce the amount of processed data, thereby improving the processing speed, while reducing the impact of interference information and increasing the accuracy of the analysis.

[0096] Step S50: at least input the target suture image into the trained suture maturity neural network to obtain the suture maturity.

[0097] In one embodiment, at least the target suture image is input into a specially trained neural network to assess the maturity of the suture.

[0098] In 2017, Angelieri F et al. proposed a maturity assessment standard for the zygomaticomaxillary suture.

[0099] In this embodiment, the suture maturity neural network is obtained by training a large number of target suture images with maturity labels, and can identify and predict different stages of suture maturity.

[0100] In this embodiment, the target suture image is a three-dimensional image. The neural network used by the suture maturity neural network is a deep learning model, such as a 3D Resnet50 neural network. The suture maturity neural network can extract important features from the input target suture image, and analyze and classify these features through the weights learned during the training process, and finally output the maturity level of the suture. In this embodiment, Mean Square Error is used as the loss function during the training process of the suture maturity neural network. The training of the neural network involves the use of a target suture image library with known maturity labels, which are manually annotated.

[0101] In this embodiment, the 3D Resnet50 neural network can be used to effectively process the target suture map of three-dimensional data and identify the spatial information therein, making the judgment more accurate. In addition, when using MSE as the loss function to predict continuous variables, such as suture maturity indicators, the square error between the predicted value and the actual value can be directly measured, which helps to optimize the prediction accuracy of the model.

[0102] In this embodiment, this step uses the previously processed target suture map and suture maturity neural network to automate the maturity assessment process. The prior processing of the target suture map not only improves the efficiency of the assessment, but also retains useful information related to the suture, ensuring the accuracy and repeatability of the results. Through the implementation of this step, the suture maturity can be quickly and accurately assessed.

[0103] Embodiment 2:

[0104] Most of the techniques in this embodiment are the same as those in Embodiment 1, except that the input in step S50 in this embodiment is not only the target suture image, but other techniques are the same as those in Embodiment 1, and will not be described in detail here:

[0105] In this embodiment, before step S50, the following steps are also included: Figure 3 As shown, the details are as follows:

[0106] Step A: Input the target suture image into the trained rotation neural network to obtain the rotation element.

[0107] In this embodiment, the rotation element is an element output by the neural network that controls the angle of rotation of the target image, such as the normal vector of the plane where the suture is located, the rotation matrix, the rotation angle, etc.

[0108] In one embodiment, the input of the trained rotation neural network is the target suture map. Because the target suture map has the predicted suture position information, the target suture map removes many factors irrelevant to the suture.

[0109] In one embodiment, the rotation element at least includes: a normal vector of the plane where the suture is located, which is obtained by step A-1 as follows:

[0110] Step A-1: ​​Input the target suture image into the trained rotation neural network to obtain the normal vector of the plane where the suture is located.

[0111] In one embodiment, the target suture image is input into a specially trained rotation neural network to obtain the normal vector of the plane where the suture is located. In this embodiment, the normal vector is used to indicate the best visual representation direction of the suture plane. In this step, the normal vector is obtained so that the suture screen reaches a preset angle with the observation plane as much as possible in the image or three-dimensional model.

[0112] In this embodiment, the plane where the suture is located refers to the section with the largest possible area of ​​the suture. The suture is a three-dimensional structure, and the plane where the suture is located refers to the suture section that doctors often use to observe, which is usually the section with the largest surface area.

[0113] In this embodiment, the rotational neural network uses deep learning technology to predict the normal vector of the suture plane by analyzing multi-level image features. Furthermore, the rotational neural network also uses Mean Square Error as a loss function.

[0114] In this embodiment, this is done to ensure that the rotated suture image is correctly oriented in subsequent processing and analysis. By obtaining the normal vector, the rotation angle of the target suture image can be adjusted so that the suture presents the best viewing angle in subsequent maturity assessment or other detailed analysis.

[0115] Step B: Rotate the target suture image at least based on the rotation element to obtain a rotated suture image.

[0116] In this embodiment, a preset observation angle is reached between the normal vector of the rotated suture image and the observation plane.

[0117] In one embodiment, the target suture image is rotated based on at least the normal vector of the plane or the rotation matrix to obtain a rotated suture image. In one embodiment, the target suture image is rotated according to the normal vector obtained in the previous step to ensure that the suture is presented at an optimal viewing angle in the final image.

[0118] In this embodiment, the specific implementation of the rotation operation includes calculating the rotation angle between the current image orientation and the ideal orientation described by the normal vector. This is achieved by calculating the angle difference between the normal vector and the observation plane, and using the vector dot multiplication formula and inverse trigonometric functions in mathematics to determine the necessary rotation angle. Then, the image is actually rotated using an image processing algorithm to complete a high-precision rotation adjustment.

[0119] In one embodiment, the angle of the observation plane is the angle of the second three-dimensional image.

[0120] In this embodiment, the rotated suture image meets the requirement of the preset observation angle. In one embodiment, a certain angle deviation is achieved between the normal vector of the suture plane and the observation plane (generally a vertical plane of the image).

[0121] Convenient image orientation is not only related to visual accuracy, but also affects the efficiency and effectiveness of image analysis algorithms in identifying and quantifying suture features. By automatically rotating the image to match the preset viewing angle, the diagnostic potential of the image is maximized, improving the accuracy and reliability of the subsequent analysis stage.

[0122] And this rotation also takes into account the training samples of the suture maturity neural network. In order to make the prediction accuracy of the suture maturity neural network higher, using the rotated suture image can maximize the sensitivity of the suture maturity neural network to the suture, thereby improving the prediction accuracy of the suture maturity neural network.

[0123] Furthermore, in another embodiment, the output of the rotation neural network may be a rotation matrix or a rotation angle is directly given, and then the target suture image is rotated based on the rotation matrix or the rotation angle to obtain a rotated suture image.

[0124] Furthermore, in this embodiment, step S50 also includes: step S501, which is as follows

[0125] Step S501: inputting a plurality of rotated suture images above and below the rotated suture image corresponding to the predicted suture position and the target suture image into a trained suture maturity neural network to obtain the suture maturity.

[0126] In this embodiment, the predicted suture position corresponds to a prediction unit. Since the second three-dimensional image is a three-dimensional image, when the prediction unit is a voxel, the predicted suture position finally obtained is a point in the three-dimensional space.

[0127] In one embodiment, further, the multiple rotational suture images located above and below the rotational suture image corresponding to the predicted suture position include: at least four adjacent rotational suture images located above the rotational suture image corresponding to the predicted suture position and at least four adjacent rotational suture images located below the rotational suture image corresponding to the predicted suture position.

[0128] After the predicted suture position is processed, a corresponding rotated suture image is obtained. There are other two-dimensional images above and below the rotated suture image, and these two-dimensional images have been rotated. In one embodiment, 31 layers of two-dimensional images above the rotated suture image are selected, and 32 layers of two-dimensional images below the rotated suture image are selected. And the total of 64 layers of two-dimensional images are input into the trained suture maturity neural network to obtain the maturity of the suture.

[0129] Embodiment 3:

[0130] Most of the techniques in this embodiment are the same as those in Embodiment 2, except for step S501 in this embodiment. Other techniques are the same as those in Embodiment 2, and will not be described in detail here.

[0131] In this implementation, step S50 includes step S501-1, which is specifically as follows:

[0132] Step S501 - 1 : inputting the rotated suture image and the segmented suture image into a trained suture maturity neural network to obtain the suture maturity.

[0133] In another implementation, step S50 includes step S501-2, which is specifically as follows:

[0134] Step S501 - 2: inputting the rotated suture image, the target suture image and the segmented suture image into the trained suture maturity neural network to obtain the suture maturity.

[0135] The difference between the above two implementations is that the input of the suture maturity neural network is different. Step S501-1 uses two types of data as input, and step S501-2 uses three types of data as input. The target suture map has been introduced before and will not be repeated here.

[0136] The method for obtaining the segmented suture map includes step B01 and step B02, which are specifically as follows:

[0137] Step B01: Obtain a two-dimensional suture image based on the suture position and the rotated suture image.

[0138] In one embodiment, a two-dimensional image of the suture position parallel to the observation plane is found, that is, the two-dimensional image of the suture is parallel to the observation plane.

[0139] Step B02: Input the two-dimensional suture image into a trained segmentation neural network to cut the suture and obtain a segmented suture image.

[0140] In one embodiment, the segmentation neural network is established using deep learning technology. In one embodiment, the segmentation neural network uses 2D U-net. The training of the segmentation neural network uses binary cross entropy loss and Dice loss as loss functions.

[0141] In this embodiment, 2D U-net is suitable for image segmentation problems because of its unique structural design, including symmetrical downsampling and upsampling paths, which can capture and retain the detailed information of the image, while restoring the contextual information of the image through jump connections, which helps to accurately segment small or complex structures. Combining binary cross entropy loss and Dice loss as loss functions, the model's performance in voxel-level classification and the shape similarity of the segmented regions can be optimized simultaneously. The binary cross entropy loss provides direct probability error feedback for the prediction of each voxel, which helps to improve the classification accuracy of the model; while the Dice loss emphasizes the overlap between the predicted area and the true area, which can help the segmentation neural network achieve higher segmentation accuracy even with a small number of positive samples. This combination of loss functions provides a powerful way to balance voxel-level accuracy and overall segmentation quality during the learning process, thereby enabling the segmentation neural network to achieve better performance in this application.

[0142] In one embodiment, the two-dimensional suture image is cut along the edge of the suture, and finally a segmented suture image is obtained.

[0143] This step provides different suture angles for the neural network to process through precise suture segmentation, which improves the subsequent maturity recognition effect.

[0144] Furthermore, step B02 also includes: using an image morphological method to eliminate small independent segmentation results in the segmented suture image, and only retaining the segmentation result with the largest volume as the final output result.

[0145] This step uses image morphology to remove small independent segmentation results in the segmented suture image, and only retains the segmentation result with the largest volume as the final output result. Image morphology is an image processing technology based on geometric structure, which extracts useful information by defining and manipulating the shape and structure in the image. Common morphological operations include dilation, erosion, opening and closing operations.

[0146] In one embodiment, first, the segmented suture image is converted into a binary image using a binarization process to separate the suture part from the background part. Then, a morphological operation (i.e., corrosion followed by dilation) is applied to remove small independent segmentation results. The opening operation can effectively remove small isolated voxel points while retaining the main structure, thereby enhancing the continuity of the target area.

[0147] After this, the volume of each independent segmentation result is calculated. The volume calculation can be achieved by counting the number of voxel points contained in the segmentation result. Then, the volumes of each segmentation result are compared, and the segmentation result with the largest volume is retained as the final output. This step ensures that only the main bone suture areas with clinical significance are retained, and irrelevant small noise areas are eliminated.

[0148] This step achieves image purification and optimization by applying morphological methods to remove small independent segmentation results. By retaining only the segmentation results with the largest volume, the accuracy of suture recognition is improved, ensuring the reliability of subsequent processing and analysis. In addition, this process reduces data redundancy, improves computational efficiency, and makes the entire system run more efficiently.

[0149] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art can understand that in order to achieve the effects of the present invention, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present invention.

[0150] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of each of the above method embodiments when executed by the processor. Among them, the computer program includes a computer program, and the computer program can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry a computer program. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electric carrier signals and telecommunication signals.

[0151] Furthermore, the present invention also provides a control device. In an embodiment of a control device according to the present invention, the control device includes a processor and a storage device, the storage device can be configured to store a program for executing the method for detecting the maturity of the zygomatico-maxillary suture in the above method embodiment, and the processor can be configured to execute the program in the storage device, which includes but is not limited to the program for executing the method for detecting the maturity of the zygomatico-maxillary suture in the above method embodiment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The control device can be a control device device formed by various electronic devices.

[0152] Furthermore, the present invention also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium can be configured to store a program for executing the maturity detection method of the zygomaticomaxillary suture of the above-mentioned method embodiment, and the program can be loaded and run by the processor to implement the above-mentioned maturity detection method of the zygomaticomaxillary suture. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0153] Further, it should be understood that since the setting of each module is only for illustrating the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only schematic.

[0154] It is understood by those skilled in the art that each module in the device can be adaptively split or combined. Such splitting or combining of specific modules will not cause the technical solution to deviate from the principle of the present invention, and therefore, the technical solutions after splitting or combining will fall within the protection scope of the present invention.

[0155] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for detecting the maturity of the zygomaticomaxillary suture, characterized in that: include: Providing a first three-dimensional image corresponding to the zygomaticomaxillary region of the subject to be examined, wherein the first three-dimensional image is obtained based on medical imaging of the zygomaticomaxillary region of the subject to be examined; Preprocessing the first three-dimensional image to obtain a second three-dimensional image; Obtaining a predicted suture position according to the second three-dimensional image and a trained suture position prediction neural network; Cropping the second three-dimensional image according to the predicted suture position to obtain a target suture image; At least the target suture image is input into a trained suture maturity neural network to obtain the suture maturity.

2. The method for detecting the maturity of the zygomaticomaxillary suture according to claim 1, characterized in that: The second three-dimensional image is composed of a plurality of prediction units, and the prediction unit includes at least one voxel, including: the predicted suture position is obtained according to the second three-dimensional image and the trained suture position prediction neural network, including: Inputting the second three-dimensional image into a trained suture position prediction neural network to obtain the reliability of the prediction unit, wherein the reliability is used to indicate the degree to which the prediction unit is a suture; The predicted suture position is obtained according to the reliability of the prediction unit.

3. The method for detecting the maturity of the zygomaticomaxillary suture according to claim 1, characterized in that: The preprocessing of the first three-dimensional image to obtain a second three-dimensional image includes: Using a preset display attribute to adjust the first three-dimensional image, at least obtaining a first channel image and a second channel image, wherein the first channel image and the second channel image have different display attributes; At least the first channel image and the second channel image are superimposed to obtain a second three-dimensional image with two channels.

4. The method for detecting the maturity of the zygomaticomaxillary suture according to claim 3, characterized in that: The display attributes include window width and window level, wherein the window width of the first channel map is greater than the window width of the second channel map, and the window level of the first channel map is less than the window level of the second channel map.

5. The method for detecting the maturity of the zygomaticomaxillary suture according to any one of claims 1 to 3, characterized in that: The step of obtaining a predicted suture position according to the reliability of the prediction unit includes: sorting the prediction units according to the confidence; Filter prediction units that meet a preset filtering condition from the prediction units to obtain a filtered prediction unit set; According to the predicted unit set, the predicted suture position is obtained.

6. The method for detecting the maturity of the zygomaticomaxillary suture according to claim 5, characterized in that: The step of obtaining the predicted suture position according to the prediction unit set includes: Obtaining a set of coordinate information of prediction units in a corresponding prediction unit set; The predicted suture position is determined according to the coordinate information set, wherein the coordinates corresponding to the predicted suture position at least include the median coordinates determined by the coordinate information set.

7. The method for detecting the maturity of the zygomaticomaxillary suture according to claim 1, characterized in that: Before at least inputting the target suture image into the trained suture maturity neural network to obtain the suture maturity, the method further includes: Inputting the target suture image into a trained rotation neural network to obtain rotation elements; The target suture image is rotated at least based on the rotation element to obtain a rotated suture image, wherein a normal vector in the rotated suture image reaches a preset observation angle with an observation plane, wherein the normal vector is a normal vector of the plane where the suture is located; The step of at least inputting the target suture image into a trained suture maturity neural network to obtain suture maturity includes: At least the target suture image and the rotated suture image are input into a trained suture maturity neural network to obtain the suture maturity.

8. The method for detecting the maturity of the zygomaticomaxillary suture according to claim 7, characterized in that: The step of at least inputting the target suture image and the rotated suture image into a trained suture maturity neural network to obtain suture maturity includes: The rotated suture image, the segmented suture image and the target suture image are input into the trained suture maturity neural network to obtain the suture maturity.

9. The method for detecting the maturity of the zygomaticomaxillary suture according to claim 8, characterized in that: The segmented suture map is obtained by the following steps: Based on the suture position and the rotated suture image, a suture two-dimensional image is obtained; The suture two-dimensional image is input into a trained segmentation neural network to cut the suture and obtain a segmented suture image.

10. A control device, comprising a processor and a storage device, wherein the storage device is suitable for storing a plurality of computer programs, characterized in that: The computer program is suitable for being loaded and run by the processor to execute the maturity detection method of the zygomaticomaxillary suture according to any one of claims 1 to 9.

11. A computer-readable storage medium storing a plurality of computer programs, characterized in that: The computer program is suitable for being loaded and run by a processor to execute the maturity detection method of the zygomaticomaxillary suture according to any one of claims 1 to 9.

Images