Alveolar bone segmentation method, device, equipment and storage medium
The pre-trained alveolar bone segmentation model is used to perform coarse and fine segmentation and image merging processing, which solves the problem of alveolar bone segmentation relying on user interaction in the existing technology and realizes fully automatic and efficient segmentation of the alveolar bone.
Patent Information
- Application Number
- CN202111514465.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-13
- Publication Date
- 2025-09-09
- Estimated Expiration
- 2041-12-13
AI Technical Summary
Existing alveolar bone segmentation methods rely on user interaction and threshold selection, resulting in poor segmentation accuracy and efficiency.
The pre-trained target lower alveolar bone and upper alveolar bone coarse segmentation model is used for coarse segmentation, combined with the pre-trained fine segmentation model for fine segmentation, and the image merging processing is performed by merging the position information of the lower alveolar bone and upper alveolar bone to achieve fully automatic segmentation.
The accuracy and efficiency of alveolar bone segmentation are improved, and specific areas of the lower and upper alveolar bones can be effectively distinguished.
Smart Images
Figure CN114187305B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to image processing technology, and in particular to an alveolar bone segmentation method, apparatus, device and storage medium. Background Art
[0002] The alveolar bone is a crucial anatomical structure in the oral cavity, comprising both the upper and lower alveolar bones. Segmenting the alveolar bone in images allows for accurate identification of its boundaries, which is crucial for visualization, maxillofacial surgery, and oral implant planning.
[0003] Currently, alveolar bone segmentation in images can be performed based on thresholding and region growing. However, this segmentation method relies on user interaction and threshold selection, resulting in poor segmentation accuracy and efficiency. Summary of the Invention
[0004] The embodiments of the present invention provide an alveolar bone segmentation method, apparatus, device and storage medium to achieve fully automatic alveolar bone segmentation and improve segmentation accuracy and efficiency.
[0005] In a first aspect, an embodiment of the present invention provides an alveolar bone segmentation method, comprising:
[0006] Obtain original alveolar bone images;
[0007] Based on the pre-trained target lower alveolar bone coarse segmentation model and the target upper alveolar bone coarse segmentation model, respectively, the original alveolar bone image is subjected to lower alveolar bone coarse segmentation and upper alveolar bone coarse segmentation, and a lower alveolar bone coarse segmentation result image and an upper alveolar bone coarse segmentation result image are determined;
[0008] Based on the pre-trained target alveolar bone fine segmentation model, performing alveolar bone fine segmentation on the lower alveolar bone coarse segmentation result image to determine the lower alveolar bone fine segmentation result image;
[0009] Based on the pre-trained target alveolar bone fine segmentation model, performing alveolar bone fine segmentation on the upper alveolar bone coarse segmentation result image to determine the upper alveolar bone fine segmentation result image;
[0010] Based on the lower alveolar bone position information and the upper alveolar bone position information in the original alveolar bone image, the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image are merged to determine a target segmentation result image with the upper alveolar bone identifier and the lower alveolar bone identifier.
[0011] In a second aspect, an embodiment of the present invention further provides an alveolar bone segmentation device, comprising:
[0012] An original alveolar bone image acquisition module is used to acquire an original alveolar bone image;
[0013] a coarse segmentation module, configured to perform coarse segmentation of the lower alveolar bone and the upper alveolar bone on the original alveolar bone image based on a pre-trained target coarse segmentation model of the lower alveolar bone and a target coarse segmentation model of the upper alveolar bone, respectively, and determine a coarse segmentation result image of the lower alveolar bone and a coarse segmentation result image of the upper alveolar bone;
[0014] A lower alveolar bone fine segmentation module is used to perform lower alveolar bone fine segmentation on the lower alveolar bone coarse segmentation result image based on a pre-trained target lower alveolar bone fine segmentation model to determine a lower alveolar bone fine segmentation result image;
[0015] An upper alveolar bone fine segmentation module is used to perform upper alveolar bone fine segmentation on the upper alveolar bone coarse segmentation result image based on a pre-trained target upper alveolar bone fine segmentation model to determine an upper alveolar bone fine segmentation result image;
[0016] The target segmentation result image determination module is used to merge the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image based on the lower alveolar bone position information and the upper alveolar bone position information in the original alveolar bone image to determine the target segmentation result image with the upper alveolar bone identifier and the lower alveolar bone identifier.
[0017] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising:
[0018] one or more processors;
[0019] a memory for storing one or more programs;
[0020] When the one or more programs are executed by the one or more processors, the one or more processors implement the alveolar bone segmentation method provided by any embodiment of the present invention.
[0021] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the alveolar bone segmentation method provided in any embodiment of the present invention.
[0022] The embodiment of the present invention performs coarse segmentation of the lower alveolar bone and the upper alveolar bone on the original alveolar bone image based on the target coarse segmentation model of the lower alveolar bone and the target coarse segmentation model of the upper alveolar bone obtained in advance, respectively, and determines the coarse segmentation result image of the lower alveolar bone and the coarse segmentation result image of the upper alveolar bone, so that the region of interest of the lower alveolar bone and the upper alveolar bone can be obtained through the coarse segmentation operation. The coarse segmentation result image of the lower alveolar bone is performed with fine segmentation of the lower alveolar bone based on the target fine segmentation model of the lower alveolar bone obtained in advance, and the lower alveolar bone fine segmentation result image is determined, thereby achieving fine segmentation of the lower alveolar bone. The coarse segmentation result image of the upper alveolar bone is performed with fine segmentation of the upper alveolar bone based on the target fine segmentation model of the upper alveolar bone obtained in advance, and the upper alveolar bone fine segmentation result image is determined, thereby achieving fine segmentation of the upper alveolar bone. By merging the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image based on the lower and upper alveolar bone position information in the original alveolar bone image, a target segmentation result image with upper and lower alveolar bone markers is determined. This allows for effective differentiation of specific regions of the lower and upper alveolar bones based on the target segmentation result image. By using the target lower alveolar bone coarse segmentation model and the target upper alveolar bone coarse segmentation model for coarse segmentation of the upper and lower alveolar bones, and using the target lower alveolar bone fine segmentation model and the target upper alveolar bone fine segmentation model for fine segmentation of the upper and lower alveolar bones, fully automatic alveolar bone segmentation is achieved, improving segmentation accuracy and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 This is a flow chart of an alveolar bone segmentation method provided in Example 1 of the present invention;
[0024] Figure 2 This is an example of an alveolar bone segmentation result involved in the first embodiment of the present invention;
[0025] Figure 3 This is a schematic structural diagram of an alveolar bone segmentation device provided in a second embodiment of the present invention;
[0026] Figure 4 This is a structural diagram of an electronic device provided in Example 3 of the present invention. DETAILED DESCRIPTION
[0027] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.
[0028] Example 1
[0029] Figure 1This is a flow chart of a method for alveolar bone segmentation provided in the first embodiment of the present invention. This embodiment is applicable to the case of segmenting the upper alveolar bone and the lower alveolar bone in an image. The method can be performed by an alveolar bone segmentation device, which can be implemented by software and / or hardware and integrated into an electronic device. Figure 1 As shown, the method specifically includes the following steps:
[0030] S110. Acquire an original alveolar bone image.
[0031] The original alveolar bone image may refer to a cone-beam computerized tomography (CBCT) image including the user's alveolar bone acquired by using a CBCT device.
[0032] S120, based on the pre-trained target lower alveolar bone coarse segmentation model and the target upper alveolar bone coarse segmentation model, respectively perform lower alveolar bone coarse segmentation and upper alveolar bone coarse segmentation on the original alveolar bone image to determine the lower alveolar bone coarse segmentation result image and the upper alveolar bone coarse segmentation result image.
[0033] Among them, the target lower alveolar bone coarse segmentation model can be a neural network model obtained by pre-training based on sample data, and used to coarsely segment the lower alveolar bone in the image. The target upper alveolar bone coarse segmentation model can be a neural network model obtained by pre-training based on sample data, and used to coarsely segment the upper alveolar bone in the image. The target lower alveolar bone coarse segmentation model and the target upper alveolar bone coarse segmentation model in this embodiment can be any three-dimensional image segmentation network model. For example, the target lower alveolar bone coarse segmentation model and the target upper alveolar bone coarse segmentation model can both be but are not limited to: 3DUnet segmentation network model or Vnet segmentation network model. The lower alveolar bone coarse segmentation result image can be used to characterize the region of interest where the lower alveolar bone is located in the original alveolar bone image. The upper alveolar bone coarse segmentation result image can be used to characterize the region of interest where the upper alveolar bone is located in the original alveolar bone image.
[0034] Specifically, the original alveolar bone image can be input into the target coarse segmentation model of the lower alveolar bone to perform coarse segmentation of the lower alveolar bone, and based on the output of the target coarse segmentation model of the lower alveolar bone, a coarse segmentation result image of the lower alveolar bone, i.e., the region of interest of the lower alveolar bone, can be obtained. The original alveolar bone image can be input into the target coarse segmentation model of the upper alveolar bone to perform coarse segmentation of the upper alveolar bone, and based on the output of the target coarse segmentation model of the upper alveolar bone, a coarse segmentation result image of the upper alveolar bone, i.e., the region of interest of the upper alveolar bone, can be obtained.
[0035] It should be noted that this embodiment does not limit the order in which the lower and upper alveolar bone coarse segmentation operations are performed. For example, this embodiment may perform the lower alveolar bone coarse segmentation operation first and then the upper alveolar bone coarse segmentation operation; it may also perform the upper alveolar bone coarse segmentation operation first and then the lower alveolar bone coarse segmentation operation; or it may perform the lower and upper alveolar bone coarse segmentation operations simultaneously.
[0036] Exemplarily, S120 may include: preprocessing the original alveolar bone image to obtain a preprocessed target alveolar bone image; inputting the target alveolar bone image into a pre-trained target lower alveolar bone coarse segmentation model, and obtaining a lower alveolar bone coarse segmentation result image based on the output of the target lower alveolar bone coarse segmentation model; inputting the target alveolar bone image into a pre-trained target upper alveolar bone coarse segmentation model, and obtaining an upper alveolar bone coarse segmentation result image based on the output of the target upper alveolar bone coarse segmentation model.
[0037] Specifically, by preprocessing the obtained original alveolar bone image and inputting the preprocessed target alveolar bone image into the target lower alveolar bone coarse segmentation model and the target upper alveolar bone coarse segmentation model, the uniformity of the input image can be ensured, so that the model can perform alveolar bone segmentation more accurately, further improving the accuracy of segmentation.
[0038] Exemplarily, preprocessing the original alveolar bone image to obtain a preprocessed target alveolar bone image may include: normalizing the original alveolar bone image to obtain a first alveolar bone image; and cropping and resampling the first alveolar bone image to obtain a target alveolar bone image.
[0039] Specifically, the physical meaning of CT values in CBCT images, combined with the variability of different CBCT devices, can be used to statistically analyze the pixels in the original alveolar bone image. The Hounsfield Unit (HU) values in the range [0.05, 99.5] are used as the upper and lower limits for normalization. Pixel values less than the lower limit are assigned a value of 0; pixel values greater than the upper limit are assigned a value of 1. Values within the upper and lower limits are mapped to the range [0, 1] using z-score processing or linear mapping to obtain the first alveolar bone image, completing the normalization operation. By cropping and resampling the first alveolar bone image, the original size and resolution of the first alveolar bone image can be adjusted to the preset size and resolution of the input image, respectively, to ensure consistency of the input image size and resolution. For example, the preset size of the input image can be set to [128, 128, 128], and the preset resolution of the input image can be set to 1.2 mm, meaning that the actual distance corresponding to each pixel is 1.2 mm.
[0040] S130 , performing alveolar bone fine segmentation on the coarse segmentation result image of the alveolar bone based on the pre-trained target alveolar bone fine segmentation model to determine a alveolar bone fine segmentation result image.
[0041] The target alveolar bone fine segmentation model can be a neural network model pre-trained based on sample data and used to finely segment the alveolar bone in an image. In this embodiment, the target alveolar bone fine segmentation model can be any three-dimensional image segmentation network model. For example, the target alveolar bone fine segmentation model can be, but is not limited to, a 3DUnet segmentation network model or a Vnet segmentation network model. The alveolar bone fine segmentation result image can be a segmented image with a precise alveolar bone boundary location.
[0042] Specifically, the coarse segmentation result image of the lower alveolar bone, that is, the region of interest of the lower alveolar bone, can be input into the target lower alveolar bone fine segmentation model for further segmentation, and based on the output of the target lower alveolar bone fine segmentation model, the lower alveolar bone fine segmentation result image can be obtained, thereby realizing the fine segmentation of the lower alveolar bone.
[0043] Exemplarily, S130 may include: performing crop processing on the coarse segmentation result image of the lower alveolar bone to obtain an inputtable first segmentation result image; inputting the first segmentation result image into a pre-trained target lower alveolar bone fine segmentation model, and obtaining an upper alveolar bone fine segmentation result image based on the output of the target lower alveolar bone fine segmentation model.
[0044] Specifically, by cropping the coarse segmentation result image of the mandibular alveolar bone, the size of the coarse segmentation result image can be adjusted to the preset size of the input image, that is, a first segmentation result image of the preset size is obtained, thereby ensuring the uniformity of the input image size. By using the target mandibular alveolar bone fine segmentation model to segment the fixed-size first segmentation result image, the segmentation accuracy can be further improved.
[0045] S140 , performing alveolar bone fine segmentation on the upper alveolar bone coarse segmentation result image based on the pre-trained target alveolar bone fine segmentation model to determine the upper alveolar bone fine segmentation result image.
[0046] The target alveolar bone fine segmentation model can be a neural network model pre-trained based on sample data and used to finely segment the alveolar bone in an image. In this embodiment, the target alveolar bone fine segmentation model can be any three-dimensional image segmentation network model. For example, the target alveolar bone fine segmentation model can be, but is not limited to, a 3DUnet segmentation network model or a Vnet segmentation network model. The alveolar bone fine segmentation result image can refer to a segmented image with a precise alveolar bone boundary location.
[0047] Specifically, the upper alveolar bone coarse segmentation result image, that is, the region of interest of the upper alveolar bone, can be input into the target upper alveolar bone fine segmentation model for further segmentation, and based on the output of the target upper alveolar bone fine segmentation model, the upper alveolar bone fine segmentation result image can be obtained, thereby realizing the fine segmentation of the upper alveolar bone.
[0048] Exemplarily, S140 may include: performing crop processing on the coarse segmentation result image of the upper alveolar bone to obtain a second segmentation result image that can be input; inputting the first segmentation result image into a pre-trained target upper alveolar bone fine segmentation model, and obtaining the upper alveolar bone fine segmentation result image based on the output of the target upper alveolar bone fine segmentation model.
[0049] Specifically, by cropping the coarse segmentation result image of the upper alveolar bone, the coarse segmentation result image of the lower alveolar bone can be resized to the preset size of the input image, thereby obtaining a second segmentation result image of the preset size, thereby ensuring the uniformity of the input image size. By using the target lower alveolar bone fine segmentation model to segment the fixed-size second segmentation result image, segmentation accuracy can be further improved.
[0050] S150 , based on the lower alveolar bone position information and the upper alveolar bone position information in the original alveolar bone image, merge the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image to determine a target segmentation result image with the upper alveolar bone identifier and the lower alveolar bone identifier.
[0051] Specifically, the coordinate position information of each pixel point in the original coordinate system corresponding to the original alveolar bone image in the result image of the fine segmentation of the lower alveolar bone and the result image of the fine segmentation of the upper alveolar bone can be used to merge the images, and the identification of each pixel point can be made based on whether it belongs to the upper alveolar bone or the lower alveolar bone, thereby obtaining a target segmentation result image with both the upper alveolar bone identification and the lower alveolar bone identification. The upper alveolar bone identification and the lower alveolar bone identification in this embodiment can be different colors to more clearly and intuitively distinguish the lower alveolar bone from the upper alveolar bone. For example, Figure 2 An example of alveolar bone segmentation results is provided. This embodiment can directly use the target segmentation result image for visualization of the upper and lower alveolar bones, as well as for segmenting the upper and lower alveolar bone regions corresponding to the upper and lower teeth in the implant recommendation algorithm.
[0052] Exemplarily, this embodiment may further resample the target segmentation result image to restore the original resolution of the target segmentation result image and improve the user viewing experience.
[0053] Exemplarily, S150 may include: if there is an overlapping area after the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image are merged, then based on the connected domain method, determine whether each pixel point in the overlapping area belongs to the lower alveolar bone pixel point or the upper alveolar bone pixel point. Specifically, based on the connected domain where each pixel point in the overlapping area is located, it can be determined whether each pixel point belongs to the lower alveolar bone pixel point or the upper alveolar bone pixel point. For example, if the lower alveolar connected domain where the pixel point is located is larger than the upper alveolar connected domain where the pixel point is located, it can be determined that the pixel point belongs to the lower alveolar bone pixel point, and the lower alveolar bone identifier is used to characterize the pixel point, thereby achieving identification confirmation of the uniqueness of the pixel point.
[0054] The technical solution of this embodiment performs coarse segmentation of the lower alveolar bone and the upper alveolar bone on the original alveolar bone image based on the target coarse segmentation model of the lower alveolar bone and the target coarse segmentation model of the upper alveolar bone obtained in advance, respectively, and determines the coarse segmentation result image of the lower alveolar bone and the coarse segmentation result image of the upper alveolar bone, so that the regions of interest of the lower alveolar bone and the upper alveolar bone can be obtained through the coarse segmentation operation. The coarse segmentation result image of the lower alveolar bone is performed on the coarse segmentation result image of the lower alveolar bone based on the target fine segmentation model of the lower alveolar bone obtained in advance, and determines the fine segmentation result image of the lower alveolar bone, thereby achieving fine segmentation of the lower alveolar bone. The coarse segmentation result image of the upper alveolar bone is performed on the coarse segmentation result image of the upper alveolar bone based on the target fine segmentation model of the upper alveolar bone obtained in advance, and determines the fine segmentation result image of the upper alveolar bone, thereby achieving fine segmentation of the upper alveolar bone. By merging the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image based on the lower and upper alveolar bone position information in the original alveolar bone image, a target segmentation result image with upper and lower alveolar bone markers is determined. This allows for effective differentiation of specific regions of the lower and upper alveolar bones based on the target segmentation result image. By using the target lower alveolar bone coarse segmentation model and the target upper alveolar bone coarse segmentation model for coarse segmentation of the upper and lower alveolar bones, and using the target lower alveolar bone fine segmentation model and the target upper alveolar bone fine segmentation model for fine segmentation of the upper and lower alveolar bones, fully automatic alveolar bone segmentation is achieved, improving segmentation accuracy and efficiency.
[0055] Based on the above technical solution, the training process of the target mandibular alveolar bone rough segmentation model may include the following steps S210-S240:
[0056] S210 , obtaining sample data for training a rough segmentation model of the mandibular alveolar bone, the sample data including: a first sample image and a standard segmentation result image corresponding to the first sample image.
[0057] The first sample image may refer to a CBCT image of the user's alveolar bone acquired using a CBCT device, or an image obtained by preprocessing the acquired CBCT image to ensure the uniformity of the input image. The preprocessing method is described above and will not be repeated here.
[0058] The standard segmentation result image may be obtained by a professional dentist marking the specific boundary position of the alveolar bone in the first sample image.
[0059] This embodiment can increase the amount of sample data by data augmentation to compensate for the lack of sample data and improve the robustness of the model. For example, the amount of sample data can be increased by randomly rotating, randomly translating, and / or left-right mirroring the existing first sample image and the corresponding standard segmentation result image. The angle range of the random rotation can be set to [-5, 5], and the number of pixels of the random translation can be set to [-10, 10]. Through the data augmentation step, the data of the data sample can be increased to compensate for the problem of insufficient data and the robustness of the model can be strengthened.
[0060] S220: Input the first sample image into the mandibular alveolar bone rough segmentation model to be trained, and obtain a first sample segmentation result image based on the output of the mandibular alveolar bone rough segmentation model.
[0061] S230 : Based on a loss function composed of a cross entropy function and a Dice function, a training error is determined according to the first sample segmentation result image and the standard segmentation result image.
[0062] The cross entropy function may refer to a distribution-based image segmentation loss function, and the Dice function may refer to a region-based image segmentation loss function.
[0063] Specifically, the weighted sum of the cross entropy function and the Dice function can be used as the final image segmentation loss function to calculate the loss error, so as to improve the training effect.
[0064] S240, back-propagating the training error to the coarse segmentation model of the lower alveolar bone, adjusting the weights in the coarse segmentation model of the lower alveolar bone, and completing the training until a preset convergence condition is reached, thereby obtaining a target coarse segmentation model of the lower alveolar bone.
[0065] Specifically, the training error is back-propagated to the coarse segmentation model of the lower alveolar bone, and the weights in the coarse segmentation model of the lower alveolar bone are adjusted until the preset convergence conditions are reached. For example, when the training error is less than the preset error or the error variation range tends to be stable, or when the number of iterative training times is equal to the preset number, it indicates that the training of the coarse segmentation model of the lower alveolar bone is completed. At this time, the trained target coarse segmentation model of the lower alveolar bone can extract the region of interest of the lower alveolar bone in the image.
[0066] Based on the above technical solution, the training process of the target upper alveolar bone rough segmentation model is similar to the training process of the target lower alveolar bone rough segmentation model, and specifically may include the following steps S310-S340:
[0067] S310 , obtaining sample data for training a rough segmentation model of the upper alveolar bone, the sample data including: a first sample image and a standard segmentation result image corresponding to the first sample image.
[0068] The sample images used for training the upper alveolar bone coarse segmentation model and the lower alveolar bone coarse segmentation model in this embodiment may be the same.
[0069] S320: Input the first sample image into the upper alveolar bone rough segmentation model to be trained, and obtain a second sample segmentation result image based on the output of the upper alveolar bone rough segmentation model.
[0070] S330 : Based on a loss function composed of a cross entropy function and a Dice function, a training error is determined according to the second sample segmentation result image and the standard segmentation result image.
[0071] S340, back-propagating the training error to the upper alveolar bone coarse segmentation model, adjusting the weights in the upper alveolar bone coarse segmentation model, and completing the training until a preset convergence condition is reached to obtain a target upper alveolar bone coarse segmentation model.
[0072] Based on the above technical solution, the training process of the target mandibular alveolar bone segmentation model may include the following steps S410-S440:
[0073] S410 , obtaining sample data for training a mandibular alveolar bone segmentation model, the sample data including: a second sample image and a standard segmentation result image corresponding to the second sample image.
[0074] The second sample image may refer to a coarse segmentation result image obtained by the first sample image based on a target mandibular alveolar bone coarse segmentation model.
[0075] S420: Input the second sample image into the mandibular alveolar bone fine segmentation model to be trained, and obtain a third sample segmentation result image based on the output of the mandibular alveolar bone fine segmentation model.
[0076] S430 : Based on a loss function composed of a cross entropy function and a Dice function, a training error is determined according to the third sample segmentation result image and the standard segmentation result image.
[0077] S440, back-propagating the training error to the mandibular alveolar bone segmentation model, adjusting the weights in the mandibular alveolar bone segmentation model, and completing the training until a preset convergence condition is reached to obtain a target mandibular alveolar bone segmentation model.
[0078] Based on the above technical solution, the training process of the target alveolar bone fine segmentation model may include the following steps S510-S540:
[0079] S510: Acquire sample data for training a superior alveolar bone fine segmentation model, the sample data including: a third sample image and a standard segmentation result image corresponding to the third sample image.
[0080] The third sample image may refer to a coarse segmentation result image obtained by the first sample image based on the target upper alveolar bone coarse segmentation model.
[0081] S520: Input the third sample image into the upper alveolar bone fine segmentation model to be trained, and obtain a fourth sample segmentation result image based on the output of the upper alveolar bone fine segmentation model.
[0082] S530 : Determine a training error based on a loss function composed of a cross entropy function and a Dice function and according to the fourth sample segmentation result image and the standard segmentation result image.
[0083] S540 , back-propagating the training error to the upper alveolar bone fine segmentation model, adjusting the weights in the upper alveolar bone fine segmentation model, and completing the training until a preset convergence condition is reached, thereby obtaining a target upper alveolar bone fine segmentation model.
[0084] It should be noted that, by adding a coarse segmentation step, this embodiment can obtain the regions of interest of the upper alveolar bone and the lower alveolar bone respectively, so that the upper alveolar bone fine segmentation model can be trained based on the sample image of the upper alveolar bone global information, and the lower alveolar bone fine segmentation model can be trained based on the sample image of the lower alveolar bone global information, thereby improving the segmentation accuracy.
[0085] The following is an embodiment of the alveolar bone segmentation device provided in an embodiment of the present invention. The device and the alveolar bone segmentation method of the above embodiment belong to the same inventive concept. For details not fully described in the embodiment of the alveolar bone segmentation device, please refer to the embodiment of the above alveolar bone segmentation method.
[0086] Example 2
[0087] Figure 3 This is a schematic diagram of the structure of an alveolar bone segmentation device provided in the second embodiment of the present invention. This embodiment is applicable to the case of segmenting the upper alveolar bone and the lower alveolar bone in the image, such as Figure 3As shown, the device specifically includes: an original alveolar bone image acquisition module 310, a coarse segmentation module 320, a lower alveolar bone fine segmentation module 330, an upper alveolar bone fine segmentation module 340 and a target segmentation result image determination module 350.
[0088] The original alveolar bone image acquisition module 310 is used to acquire the original alveolar bone image; the coarse segmentation module 320 is used to perform coarse segmentation of the lower alveolar bone and the upper alveolar bone on the original alveolar bone image based on the pre-trained target lower alveolar bone coarse segmentation model and the target upper alveolar bone coarse segmentation model, respectively, to determine the lower alveolar bone coarse segmentation result image and the upper alveolar bone coarse segmentation result image; the lower alveolar bone fine segmentation module 330 is used to perform fine segmentation of the lower alveolar bone on the coarse segmentation result image based on the pre-trained target lower alveolar bone fine segmentation model, to determine The lower alveolar bone fine segmentation result image; the upper alveolar bone fine segmentation module 340 is used to perform upper alveolar bone fine segmentation on the upper alveolar bone coarse segmentation result image based on the pre-trained target upper alveolar bone fine segmentation model, and determine the upper alveolar bone fine segmentation result image; the target segmentation result image determination module 350 is used to merge the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image based on the lower alveolar bone position information and the upper alveolar bone position information in the original alveolar bone image, and determine the target segmentation result image with the upper alveolar bone identifier and the lower alveolar bone identifier.
[0089] Optionally, the coarse segmentation module 320 includes:
[0090] A preprocessing unit, configured to preprocess the original alveolar bone image to obtain a preprocessed target alveolar bone image;
[0091] The lower alveolar bone coarse segmentation unit is used to input the target alveolar bone image into the pre-trained target lower alveolar bone coarse segmentation model, and obtain the lower alveolar bone coarse segmentation result image based on the output of the target lower alveolar bone coarse segmentation model;
[0092] The upper alveolar bone coarse segmentation unit is used to input the target alveolar bone image into the pre-trained target upper alveolar bone coarse segmentation model, and obtain the upper alveolar bone coarse segmentation result image based on the output of the target upper alveolar bone coarse segmentation model.
[0093] Optionally, the preprocessing unit is specifically used to: perform normalization processing on the original alveolar bone image to obtain a first alveolar bone image; and perform cropping and resampling processing on the first alveolar bone image to obtain a target alveolar bone image.
[0094] Optionally, the alveolar bone fine segmentation module 330 is specifically used to: perform crop processing on the coarse segmentation result image of the alveolar bone to obtain an input first segmentation result image; input the first segmentation result image into a pre-trained target alveolar bone fine segmentation model, and obtain an upper alveolar bone fine segmentation result image based on the output of the target alveolar bone fine segmentation model.
[0095] Optionally, the target segmentation result image determination module 350 is also used to: if there is an overlapping area after the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image are merged, then based on the connected domain method, determine whether each pixel point in the overlapping area belongs to the lower alveolar bone pixel point or the upper alveolar bone pixel point.
[0096] Optionally, the device further includes: a target mandibular alveolar bone rough segmentation model training module, specifically configured to:
[0097] Obtain sample data for training a coarse segmentation model of the lower alveolar bone, the sample data including: a first sample image and a standard segmentation result image corresponding to the first sample image; input the first sample image into the coarse segmentation model of the lower alveolar bone to be trained, and obtain the first sample segmentation result image based on the output of the coarse segmentation model of the lower alveolar bone; determine the training error based on the first sample segmentation result image and the standard segmentation result image based on a loss function composed of a cross entropy function and a Dice function; backpropagate the training error to the coarse segmentation model of the lower alveolar bone, adjust the weights in the coarse segmentation model of the lower alveolar bone, and terminate the training when a preset convergence condition is reached to obtain a target coarse segmentation model of the lower alveolar bone.
[0098] Optionally, the target lower alveolar bone coarse segmentation model, the target upper alveolar bone coarse segmentation model, the target lower alveolar bone fine segmentation model and the target upper alveolar bone fine segmentation model are all: 3DUnet segmentation network model or Vnet segmentation network model.
[0099] The alveolar bone segmentation device provided in the embodiment of the present invention can execute the alveolar bone segmentation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the alveolar bone segmentation method.
[0100] It is worth noting that in the embodiment of the above-mentioned alveolar bone segmentation device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.
[0101] Example 3
[0102] Figure 4 This is a structural diagram of an electronic device provided in Example 3 of the present invention. Figure 4A block diagram of an exemplary electronic device 12 suitable for implementing embodiments of the present invention is shown. Figure 4 The electronic device 12 shown is only an example and should not limit the functionality and scope of use of the embodiments of the present invention.
[0103] like Figure 4 As shown, electronic device 12 is implemented as a general-purpose computing device. Components of electronic device 12 may include, but are not limited to, one or more processors or processing units 16, system memory 28, and a bus 18 that connects various system components (including system memory 28 and processing unit 16).
[0104] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures. Examples of these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.
[0105] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.
[0106] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be configured to read and write non-removable, non-volatile magnetic media ( Figure 4 Not shown, often called a "hard drive"). Although Figure 4 Not shown, a magnetic disk drive for reading and writing to a removable non-volatile magnetic disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. System memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of various embodiments of the present invention.
[0107] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.
[0108] The electronic device 12 can also communicate with one or more external devices 14 (e.g., a keyboard, pointing device, display 24, etc.), one or more devices that enable a user to interact with the electronic device 12, and / or any device that enables the electronic device 12 to communicate with one or more other computing devices (e.g., a network card, a modem, etc.). Such communication can occur via an input / output (I / O) interface 22. Furthermore, the electronic device 12 can communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the electronic device 12 via a bus 18. It should be understood that, although not shown, other hardware and / or software modules can be used in conjunction with the electronic device 12, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0109] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the steps of an alveolar bone segmentation method provided in an embodiment of the present invention, which includes:
[0110] Obtain original alveolar bone images;
[0111] Based on the pre-trained target lower alveolar bone rough segmentation model and the target upper alveolar bone rough segmentation model, the original alveolar bone image is subjected to lower alveolar bone rough segmentation and upper alveolar bone rough segmentation respectively, and the lower alveolar bone rough segmentation result image and the upper alveolar bone rough segmentation result image are determined;
[0112] Based on the pre-trained target alveolar bone fine segmentation model, the lower alveolar bone coarse segmentation result image is subjected to lower alveolar bone fine segmentation to determine the lower alveolar bone fine segmentation result image;
[0113] Based on the pre-trained target alveolar bone fine segmentation model, the alveolar bone coarse segmentation result image is subjected to alveolar bone fine segmentation to determine the alveolar bone fine segmentation result image;
[0114] Based on the lower alveolar bone position information and the upper alveolar bone position information in the original alveolar bone image, the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image are merged to determine the target segmentation result image with the upper alveolar bone marker and the lower alveolar bone marker.
[0115] Of course, those skilled in the art will appreciate that the processor may also implement the technical solution of the alveolar bone segmentation method provided in any embodiment of the present invention.
[0116] Example 4
[0117] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the alveolar bone segmentation method provided in any embodiment of the present invention are implemented. The method includes:
[0118] Obtain original alveolar bone images;
[0119] Based on the pre-trained target lower alveolar bone rough segmentation model and the target upper alveolar bone rough segmentation model, the original alveolar bone image is subjected to lower alveolar bone rough segmentation and upper alveolar bone rough segmentation respectively, and the lower alveolar bone rough segmentation result image and the upper alveolar bone rough segmentation result image are determined;
[0120] Based on the pre-trained target alveolar bone fine segmentation model, the lower alveolar bone coarse segmentation result image is subjected to lower alveolar bone fine segmentation to determine the lower alveolar bone fine segmentation result image;
[0121] Based on the pre-trained target alveolar bone fine segmentation model, the alveolar bone coarse segmentation result image is subjected to alveolar bone fine segmentation to determine the alveolar bone fine segmentation result image;
[0122] Based on the lower alveolar bone position information and the upper alveolar bone position information in the original alveolar bone image, the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image are merged to determine the target segmentation result image with the upper alveolar bone marker and the lower alveolar bone marker.
[0123] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device.
[0124] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.
[0125] Program code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
[0126] Computer program code for performing the operations of the present invention can be written in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., through the Internet using an Internet service provider).
[0127] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computing device. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Alternatively, they can be implemented using program code executable by a computer device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module. Thus, the present invention is not limited to any specific combination of hardware and software.
[0128] Note that the above are only preferred embodiments of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments and may include many other equivalent embodiments without departing from the concept of the present invention. The scope of the present invention is determined by the scope of the appended claims.
Claims
1. A method for segmenting an alveolar bone, characterized in that: include: Obtain original alveolar bone images; Based on the pre-trained target lower alveolar bone coarse segmentation model and the target upper alveolar bone coarse segmentation model, respectively, the original alveolar bone image is subjected to lower alveolar bone coarse segmentation and upper alveolar bone coarse segmentation, and a lower alveolar bone coarse segmentation result image and an upper alveolar bone coarse segmentation result image are determined; Based on the pre-trained target alveolar bone fine segmentation model, performing alveolar bone fine segmentation on the lower alveolar bone coarse segmentation result image to determine the lower alveolar bone fine segmentation result image; Based on the pre-trained target alveolar bone fine segmentation model, performing alveolar bone fine segmentation on the upper alveolar bone coarse segmentation result image to determine the upper alveolar bone fine segmentation result image; Based on the lower alveolar bone position information and the upper alveolar bone position information in the original alveolar bone image, merging the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image to determine a target segmentation result image with an upper alveolar bone marker and a lower alveolar bone marker; The method of performing coarse segmentation of the lower alveolar bone and the upper alveolar bone on the original alveolar bone image based on the pre-trained target coarse segmentation model of the lower alveolar bone and the target coarse segmentation model of the upper alveolar bone, respectively, and determining a coarse segmentation result image of the lower alveolar bone and the coarse segmentation result image of the upper alveolar bone, comprises: Preprocessing the original alveolar bone image to obtain a preprocessed target alveolar bone image; Inputting the target alveolar bone image into a pre-trained target lower alveolar bone rough segmentation model, and obtaining a lower alveolar bone rough segmentation result image based on the output of the target lower alveolar bone rough segmentation model; Inputting the target alveolar bone image into a pre-trained target upper alveolar bone rough segmentation model, and obtaining an upper alveolar bone rough segmentation result image based on the output of the target upper alveolar bone rough segmentation model; The preprocessing of the original alveolar bone image to obtain a preprocessed target alveolar bone image includes: performing normalization processing on the original alveolar bone image to obtain a first alveolar bone image; The first alveolar bone image is cropped and resampled to obtain a target alveolar bone image.
2. The method according to claim 1, characterized in that The method of performing alveolar bone fine segmentation on the coarse segmentation result image of the lower alveolar bone based on the pre-trained target alveolar bone fine segmentation model to determine the alveolar bone fine segmentation result image includes: Performing a crop process on the rough segmentation result image of the mandibular alveolar bone to obtain an inputtable first segmentation result image; The first segmentation result image is input into a pre-trained target mandibular alveolar bone fine segmentation model, and a mandibular alveolar bone fine segmentation result image is obtained based on an output of the target mandibular alveolar bone fine segmentation model.
3. The method according to claim 1, characterized in that The training process of the target mandibular alveolar bone rough segmentation model includes: Acquire sample data for training a coarse segmentation model of the mandibular alveolar bone, the sample data comprising: a first sample image and a standard segmentation result image corresponding to the first sample image; Inputting the first sample image into a coarse segmentation model of the lower alveolar bone to be trained, and obtaining a first sample segmentation result image based on an output of the coarse segmentation model of the lower alveolar bone; Determining a training error based on the first sample segmentation result image and the standard segmentation result image based on a loss function composed of a cross entropy function and a Dice function; The training error is back-propagated to the coarse segmentation model of the lower alveolar bone, and the weights in the coarse segmentation model of the lower alveolar bone are adjusted until the training ends when a preset convergence condition is reached, thereby obtaining a target coarse segmentation model of the lower alveolar bone.
4. The method according to any one of claims 1 to 3, characterized in that The target lower alveolar bone coarse segmentation model, the target upper alveolar bone coarse segmentation model, the target lower alveolar bone fine segmentation model and the target upper alveolar bone fine segmentation model are all: 3DUnet segmentation network model or Vnet segmentation network model.
5. An alveolar bone segmentation device, characterized in that: include: An original alveolar bone image acquisition module is used to acquire an original alveolar bone image; a coarse segmentation module, configured to perform coarse segmentation of the lower alveolar bone and the upper alveolar bone on the original alveolar bone image based on a pre-trained target coarse segmentation model of the lower alveolar bone and a pre-trained target coarse segmentation model of the upper alveolar bone, respectively, and determine a coarse segmentation result image of the lower alveolar bone and a coarse segmentation result image of the upper alveolar bone; wherein the apparatus does not limit the order in which the coarse segmentation operations of the lower alveolar bone and the coarse segmentation operations of the upper alveolar bone are performed; A lower alveolar bone fine segmentation module is used to perform lower alveolar bone fine segmentation on the lower alveolar bone coarse segmentation result image based on a pre-trained target lower alveolar bone fine segmentation model to determine a lower alveolar bone fine segmentation result image; An upper alveolar bone fine segmentation module is used to perform upper alveolar bone fine segmentation on the upper alveolar bone coarse segmentation result image based on a pre-trained target upper alveolar bone fine segmentation model to determine an upper alveolar bone fine segmentation result image; a target segmentation result image determination module, configured to merge the lower alveolar bone fine segmentation result image and the upper alveolar bone fine segmentation result image based on the lower alveolar bone position information and the upper alveolar bone position information in the original alveolar bone image, and determine a target segmentation result image having the upper alveolar bone identifier and the lower alveolar bone identifier; The coarse segmentation module includes: a preprocessing unit, configured to preprocess the original alveolar bone image to obtain a preprocessed target alveolar bone image; a lower alveolar bone coarse segmentation unit, configured to input the target alveolar bone image into a pre-trained target lower alveolar bone coarse segmentation model, and obtain a lower alveolar bone coarse segmentation result image based on an output of the target lower alveolar bone coarse segmentation model; an upper alveolar bone coarse segmentation unit, configured to input the target alveolar bone image into a pre-trained target upper alveolar bone coarse segmentation model, and obtain an upper alveolar bone coarse segmentation result image based on an output of the target upper alveolar bone coarse segmentation model; The preprocessing unit is further configured to: perform normalization processing on the original alveolar bone image to obtain a first alveolar bone image; and perform cropping and resampling processing on the first alveolar bone image to obtain a target alveolar bone image.
6. An electronic device, characterized in that: The electronic device comprises: one or more processors; a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the alveolar bone segmentation method according to any one of claims 1 to 4.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the alveolar bone segmentation method according to any one of claims 1 to 4 is implemented.