A bone segmentation method and device for ankle joint CT images

Abstract

This invention discloses a method and apparatus for bone segmentation in ankle CT images. The invention acquires the ankle CT image to be segmented and the number of feet in the image. Based on the number of feet, it sequentially uses techniques such as threshold segmentation, connected component analysis, and image addition to generate a reference skeletal structure image and a seed image containing each skeletal structure and its corresponding label value. Using the reference skeletal structure image and the seed image, it acquires noisy label images with the same parameters as the ankle CT image to be segmented. These noisy label images are then denoised and filled to obtain label images containing the corresponding label values ​​for different skeletal structures and a 3D model file for each skeletal structure. This invention not only reduces the manpower and computational requirements of the bone segmentation process but also improves accuracy and efficiency.

Description

Technical Field

[0001] The present invention relates to the field of image segmentation, and in particular to a bone segmentation method and device for ankle joint CT images. Background Art

[0002] Traditional bone segmentation technology for CT images is mainly based on threshold segmentation. Due to the large difference in density between the bone structure and the surrounding tissues (such as soft tissue and air), the bone structure absorbs X-rays more strongly, which is reflected in the image as a brighter grayscale, representing a higher HU value; while the surrounding tissues absorb X-rays more weakly, presenting a darker grayscale in the image, representing a low HU value. By setting a suitable threshold (generally 300HU), the bone structure can be separated from the surrounding tissues. However, in general CT images, in addition to the bone structure and the surrounding tissues, there are also some interfering structures such as bed boards and metal meshes. The density of these structures is close to that of the bone structure, and the bone structure cannot be separated from them using threshold segmentation. In addition, separating the bone structure from other structures is often not the ultimate goal of bone segmentation. The bone structure needs to be further subdivided into different bone structures, such as the tibia, fibula, and talus in the ankle joint. Most commonly used bone segmentation technologies are developed based on region growing, such as the existing patent technology CN116258737A, 2023.04.03 A method and system for segmenting the femur and tibia of the knee joint based on region growing, which performs two region growings (coarse segmentation and fine segmentation) and two fillings (coarse filling and fine filling) based on the region growing algorithm to complete the segmentation of the femur and tibia of the knee joint. This invention solves the problems of incomplete segmentation and adhesion in traditional methods in knee joint segmentation, but uses two time-consuming region growing algorithms and filling algorithms, and requires artificial selection of seed points. The growth conditions are not applicable to other anatomical structures except the knee joint.

[0003] At present, mainstream CT image bone segmentation technology mainly relies on deep learning methods. Such methods usually use convolutional neural networks (CNN), transformers, generative adversarial networks (GAN), etc. as components to build deep neural network models, train a large number of CT images to learn the characteristics of bone structure and achieve accurate bone segmentation. Such methods usually have high accuracy and robustness, but require a large amount of labeled data for training, and the labeled data is generally obtained by experts in related fields using traditional methods. In addition, due to the large number of model parameters, such methods have high requirements for computing resources and computer configuration, slow running speed, and weak generalization for images from different anatomical structures and different scanning machines.

[0004] Therefore, an efficient, stable, low-cost bone segmentation technology that does not require a lot of human interaction is needed to achieve the segmentation of various bone structures in ankle CT images. Summary of the invention

[0005] The present invention provides a bone segmentation method and device for ankle joint CT images, so as to solve the technical problem of how to realize automatic segmentation of tibia, fibula, talus and calcaneus in ankle joint CT images with high efficiency, high quality and low labor.

[0006] In order to solve the above technical problems, in a first aspect, an embodiment of the present invention provides a bone segmentation method for ankle joint CT images, comprising:

[0007] Acquire an ankle joint CT image to be segmented and the number of feet in the ankle joint CT image;

[0008] In combination with the number of the feet, a reference bone structure CT image and a background label image are extracted from the ankle joint CT image to be segmented by multiple threshold segmentations;

[0009] Combined with the number of the feet, a seed image including all bone structures and corresponding label values ​​is generated according to the reference bone structure CT image and the background label image through threshold segmentation, connected domain analysis and image addition operation in sequence;

[0010] According to the reference bone structure image and the seed image, a noisy label image with the same parameters as the ankle joint CT image to be segmented is obtained, and the noisy label image is denoised and filled respectively to obtain a label image containing label values ​​corresponding to different bone structures and a three-dimensional model file of each bone structure; the parameters include: size, voxel spacing and orientation information.

[0011] Compared with the prior art, the above embodiment has the following beneficial effects: through multiple threshold segmentation, combined with connected domain analysis, background clipping and image addition processing technology, invalid information in the CT image is eliminated to obtain a seed image, wherein multiple customized threshold segmentation and connected domain analysis technology ensure that the initial seed areas of each bone structure finally obtained do not cross and adhere to each other; the purpose of image addition is to fill the positions of other bone structures except the tibia, fibula, talus and calcaneus of interest as background, which can make subsequent area growth faster and avoid erroneous growth; through a series of post-processing technologies, a complete three-dimensional bone model and a clean labeled image are obtained, and the clarity of the image segmentation result is improved; the above method does not require a large number of manual annotation operations and neural network training tasks, and comprehensively reduces the labor cost, time cost and hardware cost of bone segmentation of CT images.

[0012] In an embodiment of the first aspect, combining the number of the feet, sequentially performing threshold segmentation, connected domain analysis, and image addition operations, generating a seed image including all bone structures and corresponding label values ​​according to the reference bone structure CT image and the background label image, comprises:

[0013] According to a preset first segmentation threshold interval, through a threshold segmentation operation, the CT values ​​of the pixels in the reference bone structure CT image are reset to obtain a first threshold segmentation image;

[0014] Acquire each first connected domain of the first threshold segmented image, retain a corresponding number of the first connected domains according to the number of the feet and the area of ​​the first connected domains, and set the label values ​​of the remaining first connected domains as the first labels to acquire an initial seed image;

[0015] An image addition operation is performed on the background label image and the initial seed image to obtain the seed image.

[0016] Compared with the prior art, the above embodiment has the following beneficial effects: the customized threshold segmentation and connected domain analysis technology ensures that the initial seed regions of each bone structure do not cross and adhere to each other; the purpose of image addition is to fill the positions of other bone structures except the tibia, fibula, talus and calcaneus of interest as the background, which can make the subsequent region grow faster and avoid erroneous growth; at the same time, the method can automatically generate seed images, avoid excessive labor costs and human errors, and improve the accuracy of seed image generation results.

[0017] In an embodiment of the first aspect, the extracting of a bone structure CT image and a background label image for reference from the ankle joint CT image to be segmented by multiple threshold segmentations in combination with the number of feet includes:

[0018] Based on the second segmentation threshold interval and the number of the feet, processing the ankle joint CT image to obtain a de-interferenced structural CT image;

[0019] Extracting a minimum region of interest CT image from the interference-removed structural CT image based on a first segmentation threshold;

[0020] Based on the third segmentation threshold interval, the second segmentation threshold and the number of the feet, a bone structure CT image and a background label image for reference are extracted from the minimum region of interest CT image.

[0021] Compared with the prior art, the above embodiment has the following beneficial effects: through multiple customized threshold segmentations, the noise in the ankle joint CT image to be segmented is removed one by one, thereby realizing the demand for automated noise removal, improving the noise processing efficiency while accelerating the processing speed of the subsequent bone segmentation process, and improving the integrity of the bone structure in the final retained CT image, so that the seed points in the subsequent regional growing technology have room to grow.

[0022] In an embodiment of the first aspect, the processing, based on the second segmentation threshold interval and the number of feet, to obtain a de-interferenced structural CT image from the ankle joint CT image includes:

[0023] Based on the second segmentation threshold interval, the CT value of each pixel in the ankle joint CT image is reset by using a threshold segmentation technique to obtain a second threshold segmentation image;

[0024] Acquire each second connected domain of the second threshold segmented image, and retain a corresponding number of the second connected domains according to the number of the feet and the area of ​​each second connected domain;

[0025] A blank image with the same parameters as the ankle joint CT image is established, and the CT values ​​of the pixel points in the second connected domain retained in the second threshold segmentation image are assigned to the CT values ​​of the pixel points at the same position corresponding to the blank image in sequence to obtain the interference-free structural CT image.

[0026] Compared with the prior art, the above embodiment has the following beneficial effects: through customized threshold segmentation, the influence of interfering structures such as bed boards and metal meshes is preliminarily eliminated, and then combined with the connected domain analysis technology, the influence of remaining interfering structures is accurately screened out, thereby improving the degree of automation of the process of eliminating interfering structures such as bed boards and metal meshes and the accuracy of the elimination results.

[0027] In an embodiment of the first aspect, the first connected domain and the second connected domain are retained according to the connected domain areas from large to small.

[0028] Compared with the prior art, the above embodiment has the following beneficial effects: since the areas of the connected domains corresponding to different bones are different, therefore, combined with the area characteristics of each bone, the connected domains of the bones to be segmented can be conveniently and quickly retained, thereby improving the convenience and accuracy of bone segmentation.

[0029] In an embodiment of the first aspect, extracting a minimum region of interest CT image from the interference-removed structural CT image based on a first segmentation threshold comprises:

[0030] According to the first segmentation threshold, the label values ​​of the pixels in the de-interferenced structural CT image are reset by using a threshold segmentation technique to obtain a binary label image;

[0031] According to the binary label image, label mapping information including label values ​​and corresponding bounding boxes is obtained through a label image to label mapping filter;

[0032] The region in the de-interferenced structural CT image corresponding to the non-zero label value region in the label mapping information is retained to obtain the minimum region of interest CT image.

[0033] Compared with the prior art, the above embodiment has the following beneficial effects: based on the custom threshold segmentation technology, custom cropping is implemented to remove background information in the interfering structure CT image, and only the smallest area of ​​interest is retained, which helps to reduce the computational complexity in the subsequent processing flow and speed up the bone segmentation speed.

[0034] In an embodiment of the first aspect, extracting a bone structure CT image and a background label image for reference from the minimum region of interest CT image based on the third segmentation threshold interval, the second segmentation threshold, and the number of feet includes:

[0035] According to the third segmentation threshold interval, the CT values ​​of the pixels in the minimum region of interest CT image are reset by using a threshold segmentation technique to obtain the reference bone structure CT image;

[0036] According to the second segmentation threshold and the number of the feet, the CT values ​​of the pixels in the reference bone structure CT image are reset by using a threshold segmentation technique to obtain the background label image.

[0037] Compared with the prior art, the above embodiment has the following beneficial effects: through the custom threshold segmentation technology, the bone structure in the minimum area of ​​interest is first identified and retained. Furthermore, the custom threshold segmentation technology is again used to identify the non-bone structure area and mark the background, thereby retaining the complete bone information, which is convenient for the subsequent regional growth so that the seed point has room to grow.

[0038] In an embodiment of the first aspect, resetting the CT values ​​of pixels in the reference bone structure CT image by a threshold segmentation technique according to the second segmentation threshold and the number of the feet to obtain the background label image includes:

[0039] Determine the relationship between the pixel points in the currently traversed reference bone structure CT image and the second segmentation threshold;

[0040] If the CT value of the pixel point is greater than the second segmentation threshold, the CT value of the pixel point is set to the first preset CT value; otherwise, the corresponding preset CT value is set according to the number of feet;

[0041] Get the background label image.

[0042] Compared with the prior art, the above embodiment has the following beneficial effects: combining the automatic recognition method and the number of feet, the CT value of the background is accurately set, and the accuracy and generation efficiency of the background label image are improved.

[0043] In an embodiment of the first aspect, acquiring, based on the reference bone structure image and the seed image, a noisy label image having the same parameters as the ankle joint CT image to be segmented, comprises:

[0044] Combining the reference bone structure CT image and the seed image, according to a region growing algorithm, taking the pixel point region of the seed image as the initial region, obtaining a noisy cropped image corresponding to different label values ​​for each bone structure;

[0045] The noisy cropped image is resampled to obtain a noisy labeled image having the same parameters as the ankle joint CT image to be segmented.

[0046] Compared with the prior art, the above embodiment has the following beneficial effects: through the region generation algorithm and the resampling technology, a noisy labeled image containing the bone structure and the preliminary segmentation is obtained according to the seed image, which helps to quickly obtain the three-dimensional model of each bone and the clean labeled image later.

[0047] In an embodiment of the first aspect, denoising and filling the noisy label image respectively to obtain a label image containing label values ​​corresponding to different bone structures and a three-dimensional model file of each bone structure includes:

[0048] Extracting multiple single-label images of each label in the noisy label image one by one, and smoothing and denoising each of the single-label images by using a median filter;

[0049] Performing morphological closing operation on each of the single-label images processed by the median filter, and performing surface extraction to obtain a three-dimensional model surface of each label;

[0050] Processing the surface of each of the three-dimensional models by a surface wrapping solidification algorithm to obtain a three-dimensional model file of each bone structure;

[0051] The internal pixel points of each surface of the three-dimensional model after surface processing are filled with corresponding label values ​​to obtain a label image containing label values ​​corresponding to different bone structures.

[0052] Compared with the prior art, the above embodiment has the following beneficial effects: first, the labels in the label image are extracted one by one to obtain a single label image, so that the median filter can more effectively remove the impulse noise and salt and pepper noise generated by erroneous growth, and the morphological closing operation can more effectively remove small internal holes; further, the three-dimensional model surface is used as a template, and the internal pixel points are filled with the corresponding label values, so that a clean label image can be obtained quickly and conveniently.

[0053] In a second aspect, an embodiment of the present invention further provides a bone segmentation device for ankle CT images, comprising: an image acquisition module to be segmented, a noise removal module, a seed image generation module, and a bone segmentation result generation module;

[0054] Wherein, the to-be-segmented image acquisition module is used to acquire the to-be-segmented ankle joint CT image and the number of feet in the ankle joint CT image;

[0055] The noise removal module is used to extract the bone structure CT image and the background label image for reference from the ankle joint CT image to be segmented by multiple threshold segmentations in combination with the number of the feet;

[0056] The seed image generation module is used to combine the number of the feet, and sequentially perform threshold segmentation, connected domain analysis and image addition operations, and generate a seed image containing all bone structures and corresponding label values ​​according to the reference bone structure CT image and the background label image;

[0057] The bone segmentation result generation module is used to obtain a noisy label image with the same parameters as the ankle joint CT image to be segmented based on the reference bone structure image and the seed image, and to denoise and fill the noisy label image respectively to obtain a label image containing label values ​​corresponding to different bone structures and a three-dimensional model file of each bone structure; the parameters include: size, voxel spacing and orientation information.

[0058] In an embodiment of the second aspect, the seed image generation module includes: a first threshold segmentation unit, an initial seed image generation unit, and an image superposition unit;

[0059] The first threshold segmentation unit is used to reset the CT values ​​of the pixels in the reference bone structure CT image through a threshold segmentation operation according to a preset first segmentation threshold interval, so as to obtain a first threshold segmentation image;

[0060] The initial seed image generating unit is used to obtain each first connected domain of the first threshold segmented image, retain a corresponding number of the first connected domains according to the number of the feet and the area of ​​the first connected domains, and set the label value of the remaining first connected domains as the first label to obtain an initial seed image;

[0061] The image superposition unit is used to perform an image addition operation on the background label image and the initial seed image to obtain the seed image.

[0062] In an embodiment of the second aspect, the noise removal module includes: a first skeleton information extraction unit, a second skeleton information extraction unit and a third skeleton information extraction unit;

[0063] Wherein, the first bone information extraction unit is used to process the ankle joint CT image based on the second segmentation threshold interval and the number of the feet to obtain a de-interferenced structural CT image;

[0064] The second bone information extraction unit is used to extract a minimum region of interest CT image from the interference-free structure CT image based on the first segmentation threshold;

[0065] The third bone information extraction unit is used to extract the bone structure CT image and the background label image for reference from the minimum region of interest CT image based on the third segmentation threshold interval, the second segmentation threshold and the number of the feet.

[0066] In an embodiment of the second aspect, the first skeleton information extraction unit includes: a second threshold segmentation subunit, a second connected domain acquisition subunit, and a de-interference structure CT image acquisition subunit;

[0067] The second threshold segmentation subunit is used to reset the CT value of each pixel in the ankle joint CT image based on the second segmentation threshold interval by using the threshold segmentation technology to obtain a second threshold segmentation image;

[0068] The second connected domain acquisition subunit is used to acquire each second connected domain of the second threshold segmented image, and retain a corresponding number of the second connected domains according to the number of the feet and the area of ​​each second connected domain;

[0069] The interference-removed structural CT image acquisition subunit is used to establish a blank image with the same parameters as the ankle joint CT image, and to assign the CT values ​​of the pixel points in the second connected domain retained in the second threshold segmentation image to the CT values ​​of the pixel points at the same position corresponding to the blank image, so as to obtain the interference-removed structural CT image.

[0070] In an embodiment of the second aspect, the first connected domain and the second connected domain are retained according to the areas of the connected domains, from large to small.

[0071] In an embodiment of the second aspect, the second bone information extraction unit includes: a third threshold segmentation subunit, a label mapping subunit, and a minimum region of interest CT image acquisition subunit;

[0072] The third threshold segmentation subunit is used to reset the label values ​​of the pixels in the de-interference structure CT image by using the threshold segmentation technology according to the first segmentation threshold, so as to obtain a binary label image;

[0073] The label mapping subunit is used to obtain label mapping information including label values ​​and corresponding bounding boxes according to the binary label image through a label image to label mapping filter;

[0074] The minimum region of interest CT image acquisition subunit is used to retain the area in the interference-removed structural CT image corresponding to the non-zero label value area in the label mapping information, and acquire the minimum region of interest CT image.

[0075] In an embodiment of the second aspect, the third skeleton information extraction unit includes: a fourth threshold segmentation subunit and a fifth threshold segmentation subunit;

[0076] The fourth threshold segmentation subunit is used to reset the CT values ​​of the pixels in the minimum region of interest CT image by using the threshold segmentation technology according to the third segmentation threshold interval, so as to obtain the bone structure CT image for reference;

[0077] The fifth threshold segmentation subunit is used to reset the CT values ​​of the pixels in the reference bone structure CT image through the threshold segmentation technology according to the second segmentation threshold and the number of the feet, so as to obtain the background label image.

[0078] In an embodiment of the second aspect, the fifth threshold segmentation subunit is used to reset the CT values ​​of pixels in the reference bone structure CT image by a threshold segmentation technique according to the second segmentation threshold and the number of feet to obtain the background label image, including:

[0079] Determine the relationship between the pixel points in the currently traversed reference bone structure CT image and the second segmentation threshold;

[0080] If the CT value of the pixel point is greater than the second segmentation threshold, the CT value of the pixel point is set to the first preset CT value; otherwise, the corresponding preset CT value is set according to the number of feet;

[0081] Get the background label image.

[0082] In an embodiment of the second aspect, the bone segmentation result generation module includes a noisy label image acquisition unit, and the noisy label image acquisition unit includes: a noisy cropped image acquisition subunit and a noisy label image acquisition subunit;

[0083] The noisy cropped image acquisition subunit is used to combine the reference bone structure CT image and the seed image, and according to the region growing algorithm, use the pixel point region of the seed image as the initial region to acquire the noisy cropped image corresponding to different label values ​​for each bone structure;

[0084] The noisy labeled image acquisition subunit is used to resample the noisy cropped image to obtain a noisy labeled image with the same parameters as the ankle joint CT image to be segmented.

[0085] In an embodiment of the second aspect, the bone segmentation result generation module further includes an image post-processing unit, the image post-processing unit including: an image smoothing and denoising subunit, a three-dimensional model surface extraction subunit, a three-dimensional surface solidification subunit, and a three-dimensional model filling subunit;

[0086] The image smoothing and denoising subunit is used to extract multiple single-label images of each label in the noisy label image one by one, and smooth and denoise each of the single-label images through a median filter;

[0087] The three-dimensional model surface extraction subunit is used to perform morphological closing operation processing on each of the single-label images processed by the median filter, and perform surface extraction to obtain the three-dimensional model surface of each label;

[0088] The three-dimensional surface solidification subunit is used to process the surface of each three-dimensional model through a surface wrapping solidification algorithm to obtain a three-dimensional model file of each bone structure;

[0089] The three-dimensional model filling subunit is used to fill the internal pixel points of each surface of the three-dimensional model after surface processing with corresponding label values, and obtain a label image containing label values ​​corresponding to different bone structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 A schematic flow chart of a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0091] Figure 2 An ankle joint CT image to be segmented that is segmented by a bone segmentation method for an ankle joint CT image provided by an embodiment of the present invention;

[0092] Figure 3 Another schematic flow chart of a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0093] Figure 4 A schematic diagram of a flow chart of extracting a de-interferenced structural CT image in a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0094] Figure 5 A de-interferenced structural CT image obtained by a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0095] Figure 6 A minimum region of interest CT image obtained by a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0096] Figure 7 A bone structure CT image for reference obtained by a bone segmentation method for ankle joint CT images provided by an embodiment of the present invention;

[0097] Figure 8 A background label image obtained by a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0098] Fig. 9 A schematic diagram of a process for obtaining a seed image in a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0099] Fig.10 A seed image obtained by a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0100] Fig.11 A noisy labeled image obtained by a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0101] Fig.12 A schematic diagram of a process for obtaining a label image and a three-dimensional model file in a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0102] Fig.13 A clean label image and a three-dimensional model obtained by a bone segmentation method for ankle CT images provided by an embodiment of the present invention;

[0103] Fig.14 A schematic diagram of the structure of a bone segmentation device for ankle joint CT images provided by an embodiment of the present invention;

[0104] Fig.15A schematic diagram of the structure of a noise removal module in a bone segmentation device for ankle CT images provided by an embodiment of the present invention;

[0105] Fig.16 A schematic diagram of the structure of a seed image generation module in a bone segmentation device for ankle CT images provided by an embodiment of the present invention;

[0106] Fig.17 A schematic structural diagram of a bone segmentation result generating module in a bone segmentation device for ankle CT images provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0107] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0108] Embodiment 1

[0109] Please refer to Figure 1 , a bone segmentation method for ankle joint CT images provided by an embodiment of the present invention, includes S101 to S104, specifically:

[0110] S101: Acquire an ankle joint CT image to be segmented and the number of feet in the ankle joint CT image.

[0111] Preferably, reference Figure 2 , in order to obtain an ankle CT image, the number of feet in the ankle CT image can be input through manual recognition, or the image recognition technology can be used to detect the foot targets in the ankle CT image to realize the demand of automatically determining the number of feet.

[0112] S102: extracting a reference bone structure CT image and a background label image from the ankle joint CT image to be segmented by multiple threshold segmentations in combination with the number of the feet.

[0113] In one embodiment, the bone structure CT image and the background label image for reference are extracted from the ankle joint CT image to be segmented through multiple threshold segmentations in combination with the number of feet, including: based on the second segmentation threshold interval and the number of feet, processing the ankle joint CT image to obtain a de-interferenced structural CT image; based on the first segmentation threshold, extracting a minimum region of interest CT image from the de-interferenced structural CT image; based on the third segmentation threshold interval, the second segmentation threshold and the number of feet, extracting a bone structure CT image and a background label image for reference from the minimum region of interest CT image.

[0114] As one of the preferred embodiments, refer to Figure 3 The above embodiment and step S102 can be implemented by the following preferred implementation method:

[0115] Step A: Use threshold segmentation and connected domain analysis techniques to remove interfering structures such as bed boards and metal meshes in the input image;

[0116] Step B: cropping the interference-free structural CT image to remove the background and obtain a CT image with the smallest region of interest, in order to speed up the post-processing operation;

[0117] Step C: Perform threshold segmentation on the CT image of the region of interest, remove the tissue around the bone structure, and obtain the bone structure CT image and background label image for reference.

[0118] In one embodiment, the processing based on the second segmentation threshold interval and the number of feet to obtain a de-interferenced structural CT image from the ankle joint CT image includes: based on the second segmentation threshold interval, using the threshold segmentation technology, resetting the CT value of each pixel in the ankle joint CT image to obtain a second threshold segmented image; obtaining each second connected domain of the second threshold segmented image, and retaining a corresponding number of second connected domains according to the number of feet and the area of ​​each second connected domain; establishing a blank image with the same parameters as the ankle joint CT image, and assigning the CT values ​​of the pixel points in the second connected domain retained in the second threshold segmented image to the CT values ​​of the pixel points at the same position corresponding to the blank image in sequence, to obtain the de-interferenced structural CT image.

[0119] As one of the preferred embodiments, refer to Figure 4 The above embodiment and step A can be implemented by the following preferred implementation to obtain Figure 5 The CT image with the interfering structure removed is shown below:

[0120] Step A1: The above threshold segmentation technique is: set the second segmentation threshold interval to [0, 2500], for the CT value of each pixel in the ankle CT image, if the CT value is within the threshold range, then keep the CT value of the point, if the CT value is not within the threshold range, then set the CT value of the point to 0. The purpose of this step is to remove points with negative CT values ​​in the input image, avoid erroneously classifying all pixels into the same connected domain during connected domain analysis, and remove some interference structures on the basis of preserving all bone structures;

[0121] Step A2: The above-mentioned connected domain analysis technology is as follows: find adjacent pixels in the image after threshold segmentation processing, and classify them as the same connected domain with the same label. On this basis, analyze the connected domain image with the label, sort the connected domains from large to small according to their sizes, and only retain the largest (for the ankle CT image of a single foot) or the top two (for the ankle CT images of two feet) connected domains, and assign the label value of the connected domain (regardless of 1 or 2) to 1, so as to obtain the connected domain label image after removing the interference structure;

[0122] Step A3: Create a new image with the same size, spacing and orientation information as the input image, fill all its pixel values ​​with 0, and then traverse the connected domain label image and the input image after removing the interference structure at the same time. If the pixel value of the label image at the current position is 1, assign the pixel value of the input image at this position to the same position of the new image, so that a CT image with the interference structure removed can be obtained.

[0123] The above preferred embodiment effectively removes interference structures by setting the second segmentation threshold interval to [0,2500], wherein the left threshold interval is set to 0 to remove points with negative CT values ​​in the input image, avoiding erroneous classification of all pixels into the same connected domain during connected domain analysis; the right threshold interval is set to 2500 to remove some interference structures while preserving all bone structures.

[0124] Preferably, interfering structures such as bed boards and metal meshes in the above-mentioned ankle joint CT image can also be manually removed by real-time human interaction in combination with outlining or clicking operations.

[0125] In one embodiment, the first connected domain and the second connected domain are reserved according to the connected domain areas from large to small.

[0126] In one embodiment, based on the first segmentation threshold, extracting the minimum region of interest CT image from the de-interference structure CT image includes: according to the first segmentation threshold, resetting the label values ​​of the pixels in the de-interference structure CT image by threshold segmentation technology to obtain a binary label image; according to the binary label image, obtaining label mapping information including the label value and the corresponding bounding box by converting the label image to a label mapping filter; retaining the area in the de-interference structure CT image corresponding to the non-zero label value area in the label mapping information to obtain the minimum region of interest CT image.

[0127] As one of the preferred embodiments, the above embodiment and step B can be implemented by the following preferred implementation mode to obtain: Figure 6 The minimum region of interest CT image shown: the CT image with interference structure removed is processed using threshold segmentation, and the pixels with CT values ​​greater than 0 in the image are set to 1 to obtain a binary label image that is either 0 or 1. The LabelImageToLabelMapFilter method in the open source library ITK (InsightSegmentationandRegistrationToolkit) is used to convert it into a label map (labelmap), in which each object is composed of a label value and the corresponding bounding box information; the label map mask filter LabelMapMaskImageFilter in the ITK library is used, with the above-mentioned interference-free structure CT image and label map as input, the area corresponding to the non-zero pixels in the label map is retained on the input image, and the input image is cropped according to the bounding box of the object in the input label map to remove the blank area outside the bounding box to obtain the minimum region of interest CT image.

[0128] Preferably, when cropping the CT image with the interference structure removed, in addition to the implementation of the above embodiment, manual cropping or other cropping methods can also be used to retain the smallest region of interest.

[0129] In one embodiment, the extracting of a reference skeletal structure CT image and a background label image from the minimum region of interest CT image based on the third segmentation threshold interval, the second segmentation threshold and the number of feet includes: resetting the CT values ​​of pixels in the minimum region of interest CT image by threshold segmentation technology according to the third segmentation threshold interval to obtain the reference skeletal structure CT image; resetting the CT values ​​of pixels in the reference skeletal structure CT image by threshold segmentation technology according to the second segmentation threshold and the number of feet to obtain the background label image.

[0130] Furthermore, in the above embodiment, the CT value of the pixel point in the reference bone structure CT image is reset by threshold segmentation technology according to the second segmentation threshold and the number of feet to obtain the background label image, including: judging the relationship between the pixel point in the currently traversed bone structure CT image for reference and the second segmentation threshold; if the CT value of the pixel point is greater than the second segmentation threshold, setting the CT value of the pixel point to a first preset CT value; otherwise, setting the corresponding preset CT value according to the number of feet; and obtaining the background label image.

[0131] As one of the preferred embodiments, the above embodiment and step C can be implemented by the following preferred implementation mode to obtain: Figure 7 The CT images of the bone structure shown for reference and Figure 8 Background label image shown:

[0132] Step C1: Set the third segmentation threshold interval to [300, 2500]. For the CT value of each pixel in the CT image of the region of interest, if the CT value is within the threshold range, the CT value of the point is retained; if the CT value is not within the threshold range, the CT value of the point is set to 0. The purpose of this step is to remove the tissue around the bone structure and obtain a bone structure CT image for reference.

[0133] Step C2: Set the second segmentation threshold to 0. For each pixel point in the reference bone structure CT image, if the CT value is greater than 0, set the CT value of the point to 0. If the CT value is equal to 0, set the CT value of the point to 5 (for the ankle CT image of a single foot) or 9 (for the ankle CT image of two feet) as the background label value of the regional growth seed image to obtain a background label image.

[0134] In the above embodiment, the third segmentation threshold interval is set to [300, 2500], where the CT value of the tissue around the bone structure is less than 300, while the CT value of the bone structure is greater than 300, and this value can effectively remove the tissue around the bone structure; the reason for setting the right interval of the threshold range to 2500 is to keep consistent with the range of the second threshold segmentation interval, where, due to the threshold operation in step A1, the CT values ​​of the pixels in the CT image with the interference structure removed are all less than or equal to 2500, and the maximum value is taken here, which can effectively retain all the bone structures. In addition, the above embodiment sets the second segmentation threshold to 0, which can accurately separate the background from the bone structure, thereby obtaining the correct background label image.

[0135] S103: Based on the number of the feet, threshold segmentation, connected domain analysis and image addition operations are performed in sequence to generate a seed image containing all bone structures and corresponding label values ​​according to the reference bone structure CT image and the background label image.

[0136] As one of the preferred embodiments, refer to Figure 2 Step S103 can be implemented by the following preferred implementation method:

[0137] Step D: Perform threshold segmentation, connected domain analysis, and image addition processing on the CT image of the region of interest and the background label image to obtain a seed image corresponding to a different label value for each bone structure.

[0138] In one embodiment, the combination of the number of feet, through threshold segmentation, connected domain analysis and image addition operation, generates a seed image containing all bone structures and corresponding label values ​​according to the reference bone structure CT image and the background label image, including: according to a preset first segmentation threshold interval, through a threshold segmentation operation, resetting the CT value of the pixel point in the reference bone structure CT image to obtain a first threshold segmentation image; obtaining each first connected domain of the first threshold segmentation image, retaining a corresponding number of the first connected domains according to the number of feet and the area of ​​the first connected domain, and setting the label value of the remaining first connected domain to the first label to obtain an initial seed image; performing an image addition operation on the background label image and the initial seed image to obtain the seed image.

[0139] As one of the preferred embodiments, Fig. 9 The above embodiment and step D can be implemented by the following preferred implementation to obtain Fig.10 Seed image shown:

[0140] Step D1: Set the first segmentation threshold interval to [500, 2500]. For the CT value of each pixel in the reference bone structure CT image, if the CT value is within the threshold range, the CT value of the point is retained; if the CT value is not within the threshold range, the CT value of the point is set to 0. The purpose of this step is to remove the tissue around the bone structure while ensuring that the bone structures are not adhered to each other;

[0141] Step D2: Find adjacent pixels in the image after threshold segmentation and classify them as connected domains with the same label. Since a higher threshold is used, the bone structures are not connected to each other, so a label image corresponding to each bone structure with a different label value can be obtained. On this basis, the connected domain image with labels is analyzed and sorted from large to small according to the size of the connected domain. Only the first four (for the ankle CT image of a single foot, corresponding to the tibia, calcaneus, talus and fibula respectively) or the first eight (for the ankle CT images of two feet, corresponding to the left and right tibia, left and right calcaneus, left and right talus and left and right fibula respectively) connected domains are retained, and the label values ​​of the remaining connected domains are uniformly changed to 5 (for the ankle CT image of a single foot) or 9 (for the ankle CT images of two feet) as the background label value of the region growing seed image to obtain the initial seed image;

[0142] Step D3: Perform an image addition operation on the background label image and the initial seed image to obtain a seed image.

[0143] In the above embodiment, the first threshold segmentation interval is set to [500, 2500], wherein the left interval of the threshold range is set to 500, which can effectively remove the tissue around the bone structure, while ensuring that the various bone structures are not adhered to each other, so that the next connected domain analysis can distinguish different bone structures as different connected domains; the reason for setting the right interval of the threshold range to 2500 is to be consistent with the range of the second threshold segmentation interval. Due to the threshold operation in step A1, the CT values ​​of the pixels in the CT image with the interference structure removed are all less than or equal to 2500. Taking the maximum value here can retain all bone structures.

[0144] S104: According to the reference bone structure image and the seed image, a noisy labeled image having the same parameters as the ankle joint CT image to be segmented is obtained, and the noisy labeled image is denoised and filled respectively to obtain a label image containing label values ​​corresponding to different bone structures and a three-dimensional model file of each bone structure; the parameters include: size, voxel spacing and orientation information.

[0145] As one of the preferred embodiments, Figure 2 The above step S104 can be implemented by the following preferred implementation method:

[0146] Step E: Combine the reference bone structure image and the seed image, use region growing and resampling techniques to obtain a label image with the same size, spacing and orientation information as the input image but with noise;

[0147] Step F: Use the label image post-processing technology consisting of median filtering, morphological closing operation, three-dimensional surface reconstruction, and SurfaceWrapSolidify algorithm to denoise the noisy label image, fill holes and the spongy area inside the bone, and finally obtain a three-dimensional model of each bone structure and a clean label image.

[0148] In one embodiment, based on the reference bone structure image and the seed image, a noisy labeled image with the same parameters as the ankle joint CT image to be segmented is obtained, including: combining the reference bone structure CT image and the seed image, according to a region growing algorithm, taking the pixel point area of ​​the seed image as the initial area, and obtaining a noisy cropped image corresponding to different label values ​​for each bone structure; resampling the noisy cropped image to obtain a noisy labeled image with the same parameters as the ankle joint CT image to be segmented.

[0149] As one of the preferred embodiments, the above embodiment and step E can be implemented by the following preferred implementation mode to obtain: Fig.11 The noisy label image shown:

[0150] Step E1: Using the ITKGrowCut algorithm in the ITK library, input the reference bone structure CT image and seed image, take the pixel area of ​​the seed image as the initial area, and determine whether it is necessary to add the adjacent pixels of the initial area to the current area based on the growth condition. End the growth after the termination condition is met, and finally obtain a noisy cropped image with different label values ​​corresponding to each bone structure;

[0151] Step E2: Resample the noisy cropped image through the ResampleImageFilter method in the ITK library, set the size, spacing, and orientation information of the target image to be consistent with the input image, and use the nearest neighbor interpolation method supported by this method to interpolate to obtain a noisy labeled image with the same size, spacing, and orientation information as the original image.

[0152] Preferably, in the above step E1, other region growing algorithms may also be used, such as ITKConnectedThresholdImageFilter algorithm and floodFill algorithm in the open source library OpenCV.

[0153] Preferably, in the above step E2, a customized resampling technique is used to restore the image obtained by region growing to the same size, spacing and orientation information as the input image, or the image edges are filled to the restored size and then the spacing and orientation information are set.

[0154] In one embodiment, the noisy label image is denoised and filled respectively to obtain a label image containing label values ​​corresponding to different skeletal structures and a three-dimensional model file of each skeletal structure, including: extracting multiple single-label images of each label in the noisy label image one by one, and smoothing and denoising each of the single-label images through a median filter; performing morphological closing operation on each of the single-label images processed by the median filter, and performing surface extraction to obtain the three-dimensional model surface of each label; processing the surface of each of the three-dimensional models through a surface wrapping solidification algorithm to obtain a three-dimensional model file of each skeletal structure; filling the internal pixel points of the surface of each of the three-dimensional models after surface processing with corresponding label values ​​to obtain a label image containing label values ​​corresponding to different skeletal structures.

[0155] As one of the preferred embodiments, Fig.12 The above embodiment and step F can be implemented by the following preferred implementation to obtain Fig.13 The clean label images and 3D model files are shown:

[0156] Step F1: extract each label in the noisy label image one by one to obtain multiple single label images;

[0157] Step F2: Median filtering: Use a median filter to smooth and denoise the single-label image, removing impulse noise and salt and pepper noise while retaining the image edge information;

[0158] Step F3: Morphological closing operation: Perform morphological closing operation on the single-label image after median filtering, that is, dilation first and then erosion, in order to remove small holes in the label image;

[0159] Step F4: 3D surface reconstruction: perform surface extraction on the label image after morphological closing operation to obtain the 3D model surface of each label;

[0160] Step F5: SurfaceWrapSolidify algorithm: Process the surface of the 3D model and create a larger closed surface to wrap the surface of the 3D model to be processed. It iteratively shrinks the large closed surface to be close to the surface of the 3D model to be processed, so as to fill the holes on the surface of the 3D model. Finally, the 3D model surface can be output as a 3D model file, or it can be used as a template to fill the internal pixels with corresponding label values ​​to obtain a clean label image with the internal loose area filled.

[0161] In summary, the bone segmentation method for ankle CT images provided by the embodiment of the present invention has the following beneficial effects compared with the prior art:

[0162] 1. This embodiment uses a custom threshold segmentation and connected domain analysis technique to remove interfering structures such as bed boards and metal meshes from the segmented ankle CT images, so that subsequent processing can better focus on the bone structure; in addition, this method can automatically remove interfering structures while ensuring the correctness of the results.

[0163] 2. This embodiment uses a custom cropping technique to process the interference-free structural CT image to retain the smallest region of interest, which can reduce the number of traversals in the subsequent processing flow and speed up the operation.

[0164] 3. This embodiment uses a custom threshold segmentation technique to process the CT image of the region of interest to retain all bone structures in the image, and further processes it with a custom threshold segmentation technique to obtain a background label image. This method can completely retain the bone structure in the CT so that the seed points in the subsequent region growing technique have room to grow.

[0165] 4. This embodiment uses customized threshold segmentation, connected domain analysis and image addition technology to process the CT image of the region of interest and the background label image to obtain a seed image, wherein the customized threshold segmentation and connected domain analysis technology ensure that the initial seed regions of each bone structure do not cross and adhere to each other; image addition can fill the positions of other bone structures except the tibia, fibula, talus and calcaneus of interest as the background, thereby accelerating the subsequent regional growth speed and avoiding erroneous growth; in addition, this method is a new and fully automatic seed image generation method. The existing methods for obtaining seed images either rely on manual annotation, which is time-consuming and labor-intensive, or obtain models by training deep neural networks, which requires a large amount of data, the same time-consuming and labor-intensive annotation costs and high-cost computing resources. In comparison, the above method can quickly obtain accurate results, and the generated seed region is large, which greatly facilitates subsequent regional growth.

[0166] 5. The region growing technology in this embodiment uses the open source ITKGrowCut algorithm. The advantage of this algorithm is that it accelerates the speed of region growing by reformulating the original problem of product maximization as a problem of sum minimization. At the same time, the algorithm also supports the growth of multi-label seed images, which is very suitable for the segmentation of multiple bone structures in ankle CT images.

[0167] 6. This embodiment uses a custom resampling technology to restore the image obtained by region growing to the same size, spacing and orientation information as the input image. Since only pixels outside the region of interest need to be filled, nearest neighbor interpolation is used. Compared with other interpolation techniques, this technology is simple and fast, and does not introduce new pixel values, thereby retaining the original information of the image.

[0168] 7. This embodiment uses a customized label image post-processing technology to process the noisy label image obtained by regional growing to obtain a clean label image and a three-dimensional model of each bone structure, wherein the labels in the label image are first extracted one by one to obtain a single label image, so that the median filter can more effectively remove the pulse noise and salt and pepper noise generated by erroneous growth, and the morphological closing operation can more effectively remove small internal holes; then the label image is converted into a three-dimensional model surface, and the SurfaceWrapSolidify algorithm is used to remove the holes on the model surface. The advantage of this algorithm is that it can quickly and effectively fill the holes on the three-dimensional model surface. Further, the three-dimensional model surface is used as a template to fill the internal pixel points with the corresponding label values, so that a clean label image can be obtained after filling the internal spongy area. The above post-processing process is fast and effective.

[0169] Embodiment 2

[0170] Combination Fig.14 The embodiment of the present invention further provides a bone segmentation device for ankle CT images, comprising: an image acquisition module 11 for segmentation, a noise removal module 12, a seed image generation module 13 and a bone segmentation result generation module 14; wherein the image acquisition module 11 for segmentation is used to acquire the ankle CT image to be segmented and the number of feet in the ankle CT image; the noise removal module 12 is used to extract a bone structure CT image and a background label image for reference from the ankle CT image to be segmented by multiple threshold segmentations in combination with the number of feet; the seed image generation module 13 is used to extract a bone structure CT image and a background label image for reference in combination with the number of feet , sequentially through threshold segmentation, connected domain analysis and image addition operations, a seed image containing all bone structures and corresponding label values ​​is generated according to the reference bone structure CT image and the background label image; the bone segmentation result generation module 14 is used to obtain a noisy label image with the same parameters as the ankle joint CT image to be segmented according to the reference bone structure image and the seed image, and denoise and fill the noisy label image respectively to obtain a label image containing corresponding label values ​​of different bone structures and a three-dimensional model file of each bone structure; the parameters include: size, voxel spacing and orientation information.

[0171] refer to Fig.15, is a structural schematic diagram of the noise removal module in a bone segmentation device for ankle CT images provided in an embodiment of the present invention, comprising: a first bone information extraction unit 121, a second bone information extraction unit 122 and a third bone information extraction unit 123; wherein the first bone information extraction unit 121 is used to process the ankle CT image based on the second segmentation threshold interval and the number of feet to obtain a de-interferenced structural CT image; the second bone information extraction unit 122 is used to extract a minimum region of interest CT image from the de-interferenced structural CT image based on the first segmentation threshold; the third bone information extraction unit 123 is used to extract a reference bone structure CT image and a background label image from the minimum region of interest CT image based on the third segmentation threshold interval, the second segmentation threshold and the number of feet.

[0172] Furthermore, the first bone information extraction unit 121 is used to process the ankle joint CT image based on the second segmentation threshold interval and the number of feet to obtain a de-interferenced structural CT image, including: based on the second segmentation threshold interval, resetting the CT value of each pixel in the ankle joint CT image through the threshold segmentation technology to obtain a second threshold segmented image; obtaining each second connected domain of the second threshold segmented image, and retaining a corresponding number of second connected domains according to the number of feet and the area of ​​each second connected domain; establishing a blank image with the same parameters as the ankle joint CT image, and assigning the CT values ​​of the pixel points in the second connected domain retained in the second threshold segmented image to the CT values ​​of the pixel points at the same position corresponding to the blank image in turn, to obtain the de-interferenced structural CT image.

[0173] Furthermore, the second bone information extraction unit 122 is used to extract a minimum area of ​​interest CT image from the de-interferenced structure CT image based on a first segmentation threshold, including: according to the first segmentation threshold, by using a threshold segmentation technique, resetting the label values ​​of the pixels in the de-interferenced structure CT image to obtain a binary label image; according to the binary label image, by using a label image to label mapping filter, obtaining label mapping information including label values ​​and corresponding bounding boxes; retaining the area in the de-interferenced structure CT image corresponding to the non-zero label value area in the label mapping information to obtain the minimum area of ​​interest CT image.

[0174] Furthermore, the third bone information extraction unit 123 is used to extract a reference bone structure CT image and a background label image from the minimum area of ​​interest CT image based on the third segmentation threshold interval, the second segmentation threshold and the number of feet, including: according to the third segmentation threshold interval, by using the threshold segmentation technology, resetting the CT values ​​of the pixels in the minimum area of ​​interest CT image to obtain the reference bone structure CT image; according to the second segmentation threshold and the number of feet, by using the threshold segmentation technology, resetting the CT values ​​of the pixels in the reference bone structure CT image to obtain the background label image.

[0175] Furthermore, the CT value of the pixel in the reference bone structure CT image is reset by threshold segmentation technology according to the second segmentation threshold and the number of feet to obtain the background label image, including: judging the relationship between the pixel in the reference bone structure CT image currently traversed and the second segmentation threshold; if the CT value of the pixel is greater than the second segmentation threshold, setting the CT value of the pixel to a first preset CT value; otherwise, setting the corresponding preset CT value according to the number of feet; and obtaining the background label image.

[0176] refer to Fig.16 , is a structural schematic diagram of the seed image generation module in a bone segmentation device for ankle CT images provided by an embodiment of the present invention, comprising: a first threshold segmentation unit 131, an initial seed image generation unit 132 and an image superposition unit 133; wherein the first threshold segmentation unit 131 is used to reset the CT values ​​of the pixels in the reference bone structure CT image through a threshold segmentation operation according to a preset first segmentation threshold interval to obtain a first threshold segmented image; the initial seed image generation unit 132 is used to obtain each first connected domain of the first threshold segmented image, retain a corresponding number of the first connected domains according to the number of the feet and the area of ​​the first connected domains, and set the label value of the remaining first connected domains to the first label to obtain an initial seed image; the image superposition unit 133 is used to perform an image addition operation on the background label image and the initial seed image to obtain the seed image.

[0177] Furthermore, the first connected domain and the second connected domain are retained from large to small according to the areas of the connected domains.

[0178] refer to Fig.17 , is a structural schematic diagram of a bone segmentation result generation module in a bone segmentation device for ankle CT images provided in an embodiment of the present invention, comprising: a noisy label image acquisition unit 141 and an image post-processing unit 142.

[0179] Furthermore, the noisy labeled image acquisition unit 141 is used to obtain a noisy labeled image with the same parameters as the ankle joint CT image to be segmented based on the reference bone structure image and the seed image, including: combining the reference bone structure CT image and the seed image, according to the region growing algorithm, taking the pixel point area of ​​the seed image as the initial area, to obtain a noisy cropped image corresponding to different label values ​​for each bone structure; resampling the noisy cropped image to obtain a noisy labeled image with the same parameters as the ankle joint CT image to be segmented.

[0180] Furthermore, the image post-processing unit 142 is used to denoise and fill the noisy label image respectively, and obtain a label image containing label values ​​corresponding to different skeletal structures and a three-dimensional model file of each skeletal structure, including: extracting multiple single-label images of each label in the noisy label image one by one, and smoothing and denoising each of the single-label images through a median filter; performing morphological closing operation processing on each of the single-label images processed by the median filter, and performing surface extraction to obtain the three-dimensional model surface of each label; processing the surface of each of the three-dimensional models through a surface wrapping solidification algorithm to obtain a three-dimensional model file of each skeletal structure; filling the internal pixel points of the surface of each of the three-dimensional models after surface processing with corresponding label values ​​to obtain a label image containing label values ​​corresponding to different skeletal structures.

[0181] In summary, a bone segmentation device for ankle CT images provided in an embodiment of the present application has the following beneficial effects: through multiple threshold segmentations, combined with connected domain analysis, background clipping and image addition processing technology, invalid information in the CT image is eliminated to obtain a seed image, wherein multiple customized threshold segmentations and connected domain analysis technologies ensure that the initial seed regions of each bone structure finally obtained do not cross-adhere to each other; the purpose of image addition is to fill the positions of other bone structures except the tibia, fibula, talus and calcaneus of interest as the background, which can make subsequent areas grow faster and avoid erroneous growth; through a series of post-processing techniques, a complete three-dimensional bone model and a clean labeled image are obtained, thereby improving the clarity of the image segmentation results; the above method does not require a large number of manual annotation operations and neural network training tasks, and comprehensively reduces the labor cost, time cost and hardware cost of bone segmentation of CT images.

[0182] Embodiment 3

[0183] Based on the above-mentioned embodiment of the bone segmentation method for ankle CT images, another embodiment of the present invention provides a bone segmentation terminal device for ankle CT images. The bone segmentation terminal device for ankle CT images includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the bone segmentation method for ankle CT images of any embodiment of the present invention is implemented.

[0184] Exemplarily, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the bone segmentation device for ankle CT images.

[0185] The bone segmentation device for ankle CT images can be a computing device such as a desktop computer, a notebook, a palmtop computer, a cloud server, etc. The bone segmentation terminal device for ankle CT images can include, but is not limited to, a processor and a memory.

[0186] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The processor is the control center of the bone segmentation device for ankle CT images, and uses various interfaces and lines to connect the various parts of the entire bone segmentation device for ankle CT images. The memory may be used to store the computer program and / or module, and the processor implements various functions of the bone segmentation device for ankle CT images by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory and may also include non-volatile memory, such as a hard disk, an internal memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0187] Embodiment 4

[0188] Based on the above-mentioned embodiment of the bone segmentation method for ankle joint CT images, another embodiment of the present invention provides a storage medium, wherein the storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the bone segmentation method for ankle joint CT images of any embodiment of the present invention.

[0189] In this embodiment, the storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may 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, computer-readable media do not include electric carrier signals and telecommunication signals.

[0190] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A bone segmentation method for ankle CT images, characterized in that: include: Acquire an ankle joint CT image to be segmented and the number of feet in the ankle joint CT image; In combination with the number of the feet, a reference bone structure CT image and a background label image are extracted from the ankle joint CT image to be segmented by multiple threshold segmentations; Combined with the number of the feet, a seed image including all bone structures and corresponding label values ​​is generated according to the reference bone structure CT image and the background label image through threshold segmentation, connected domain analysis and image addition operation in sequence; According to the reference bone structure image and the seed image, a noisy label image with the same parameters as the ankle joint CT image to be segmented is obtained, and the noisy label image is denoised and filled respectively to obtain a label image containing label values ​​corresponding to different bone structures and a three-dimensional model file of each bone structure; the parameters include: size, voxel spacing and orientation information.

2. A bone segmentation method for ankle joint CT images as claimed in claim 1, characterized in that: The method combines the number of the feet, sequentially performs threshold segmentation, connected domain analysis, and image addition operations, and generates a seed image including all bone structures and corresponding label values ​​according to the reference bone structure CT image and the background label image, including: According to a preset first segmentation threshold interval, through a threshold segmentation operation, the CT values ​​of the pixels in the reference bone structure CT image are reset to obtain a first threshold segmentation image; Acquire each first connected domain of the first threshold segmented image, retain a corresponding number of the first connected domains according to the number of the feet and the area of ​​the first connected domains, and set the label values ​​of the remaining first connected domains as the first labels to acquire an initial seed image; An image addition operation is performed on the background label image and the initial seed image to obtain the seed image.

3. A bone segmentation method for ankle CT images as claimed in claim 2, characterized in that: The method of extracting a reference bone structure CT image and a background label image from the ankle joint CT image to be segmented by combining the number of the feet and performing multiple threshold segmentations includes: Based on the second segmentation threshold interval and the number of the feet, processing the ankle joint CT image to obtain a de-interferenced structural CT image; Extracting a minimum region of interest CT image from the interference-removed structural CT image based on a first segmentation threshold; Based on the third segmentation threshold interval, the second segmentation threshold and the number of the feet, a bone structure CT image and a background label image for reference are extracted from the minimum region of interest CT image.

4. A bone segmentation method for ankle joint CT images as claimed in claim 3, characterized in that: The step of processing the ankle joint CT image based on the second segmentation threshold interval and the number of feet to obtain a de-interferenced structural CT image includes: Based on the second segmentation threshold interval, the CT value of each pixel in the ankle joint CT image is reset by using a threshold segmentation technique to obtain a second threshold segmentation image; Acquire each second connected domain of the second threshold segmented image, and retain a corresponding number of the second connected domains according to the number of the feet and the area of ​​each second connected domain; A blank image with the same parameters as the ankle joint CT image is established, and the CT values ​​of the pixel points in the second connected domain retained in the second threshold segmentation image are assigned to the CT values ​​of the pixel points at the same position corresponding to the blank image in sequence to obtain the interference-free structural CT image.

5. A bone segmentation method for ankle joint CT images as claimed in claim 4, characterized in that: The first connected domain and the second connected domain are reserved from large to small according to the areas of the connected domains.

6. A bone segmentation method for ankle joint CT images as claimed in claim 3, characterized in that: The step of extracting a minimum region of interest CT image from the interference-removed structural CT image based on the first segmentation threshold comprises: According to the first segmentation threshold, the label values ​​of the pixels in the de-interferenced structural CT image are reset by using a threshold segmentation technique to obtain a binary label image; According to the binary label image, label mapping information including label values ​​and corresponding bounding boxes is obtained through a label image to label mapping filter; The region in the de-interferenced structural CT image corresponding to the non-zero label value region in the label mapping information is retained to obtain the minimum region of interest CT image.

7. A bone segmentation method for ankle joint CT images as claimed in claim 3, characterized in that: The step of extracting a reference bone structure CT image and a background label image from the minimum region of interest CT image based on the third segmentation threshold interval, the second segmentation threshold, and the number of the feet includes: According to the third segmentation threshold interval, the CT values ​​of the pixels in the minimum region of interest CT image are reset by using a threshold segmentation technique to obtain the reference bone structure CT image; According to the second segmentation threshold and the number of the feet, the CT values ​​of the pixels in the reference bone structure CT image are reset by using a threshold segmentation technique to obtain the background label image.

8. A bone segmentation method for ankle joint CT images as claimed in claim 7, characterized in that: The method of resetting the CT values ​​of the pixels in the reference bone structure CT image by using a threshold segmentation technique according to the second segmentation threshold and the number of the feet to obtain the background label image includes: Determine the relationship between the pixel points in the currently traversed reference bone structure CT image and the second segmentation threshold; If the CT value of the pixel point is greater than the second segmentation threshold, the CT value of the pixel point is set to the first preset CT value; otherwise, the corresponding preset CT value is set according to the number of feet; Get the background label image.

9. The bone segmentation method for ankle joint CT images according to claim 1, characterized in that: The step of acquiring a noisy label image having the same parameters as the ankle joint CT image to be segmented according to the reference bone structure image and the seed image comprises: Combining the reference bone structure CT image and the seed image, according to a region growing algorithm, taking the pixel point region of the seed image as the initial region, obtaining a noisy cropped image corresponding to different label values ​​for each bone structure; The noisy cropped image is resampled to obtain a noisy labeled image having the same parameters as the ankle joint CT image to be segmented.

10. The bone segmentation method for ankle joint CT images according to claim 1, characterized in that: The step of denoising and filling the noisy label image to obtain a label image containing label values ​​corresponding to different bone structures and a three-dimensional model file of each bone structure includes: Extracting multiple single-label images of each label in the noisy label image one by one, and smoothing and denoising each of the single-label images by using a median filter; Performing morphological closing operation on each of the single-label images processed by the median filter, and performing surface extraction to obtain a three-dimensional model surface of each label; Process the surface of each of the three-dimensional models by using a surface wrapping solidification algorithm to obtain a three-dimensional model file of each bone structure; The internal pixel points of each surface of the three-dimensional model after surface processing are filled with corresponding label values ​​to obtain a label image containing label values ​​corresponding to different bone structures.

11. A bone segmentation device for ankle joint CT images, characterized in that: include: An image acquisition module to be segmented, a noise removal module, a seed image generation module, and a bone segmentation result generation module; Wherein, the to-be-segmented image acquisition module is used to acquire the to-be-segmented ankle joint CT image and the number of feet in the ankle joint CT image; The noise removal module is used to extract the bone structure CT image and the background label image for reference from the ankle joint CT image to be segmented by multiple threshold segmentations in combination with the number of the feet; The seed image generation module is used to combine the number of the feet, and sequentially perform threshold segmentation, connected domain analysis and image addition operations, and generate a seed image containing all bone structures and corresponding label values ​​according to the reference bone structure CT image and the background label image; The bone segmentation result generation module is used to obtain a noisy label image with the same parameters as the ankle joint CT image to be segmented based on the reference bone structure image and the seed image, and to denoise and fill the noisy label image respectively to obtain a label image containing label values ​​corresponding to different bone structures and a three-dimensional model file of each bone structure; the parameters include: size, voxel spacing and orientation information.

12. The bone segmentation device for ankle joint CT images according to claim 11, characterized in that: The seed image generation module comprises: a first threshold segmentation unit, an initial seed image generation unit and an image superposition unit; The first threshold segmentation unit is used to reset the CT values ​​of the pixels in the reference bone structure CT image through a threshold segmentation operation according to a preset first segmentation threshold interval, so as to obtain a first threshold segmentation image; The initial seed image generating unit is used to obtain each first connected domain of the first threshold segmented image, retain a corresponding number of the first connected domains according to the number of the feet and the area of ​​the first connected domains, and set the label value of the remaining first connected domains as the first label to obtain an initial seed image; The image superposition unit is used to perform an image addition operation on the background label image and the initial seed image to obtain the seed image.

13. The bone segmentation device for ankle joint CT images according to claim 12, characterized in that: The noise removal module includes: a first skeleton information extraction unit, a second skeleton information extraction unit and a third skeleton information extraction unit; Wherein, the first bone information extraction unit is used to process the ankle joint CT image based on the second segmentation threshold interval and the number of the feet to obtain a de-interferenced structural CT image; The second bone information extraction unit is used to extract a minimum region of interest CT image from the interference-free structure CT image based on a first segmentation threshold; The third bone information extraction unit is used to extract the bone structure CT image and the background label image for reference from the minimum region of interest CT image based on the third segmentation threshold interval, the second segmentation threshold and the number of the feet.

14. The bone segmentation device for ankle joint CT images according to claim 13, characterized in that: The first bone information extraction unit is used to process the ankle joint CT image based on the second segmentation threshold interval and the number of the feet to obtain a de-interferenced structural CT image, including: Based on the second segmentation threshold interval, the CT value of each pixel in the ankle joint CT image is reset by using a threshold segmentation technique to obtain a second threshold segmentation image; Acquire each second connected domain of the second threshold segmented image, and retain a corresponding number of the second connected domains according to the number of the feet and the area of ​​each second connected domain; A blank image with the same parameters as the ankle joint CT image is established, and the CT values ​​of the pixel points in the second connected domain retained in the second threshold segmentation image are assigned to the CT values ​​of the pixel points at the same position corresponding to the blank image in sequence to obtain the interference-free structural CT image.

15. The bone segmentation device for ankle joint CT images according to claim 14, characterized in that: The first connected domain and the second connected domain are reserved from large to small according to the areas of the connected domains.

16. The bone segmentation device for ankle joint CT images according to claim 13, characterized in that: The second bone information extraction unit is used to extract a minimum region of interest CT image from the interference-free structure CT image based on a first segmentation threshold, including: According to the first segmentation threshold, the label values ​​of the pixels in the de-interferenced structural CT image are reset by using a threshold segmentation technique to obtain a binary label image; According to the binary label image, label mapping information including label values ​​and corresponding bounding boxes is obtained through a label image to label mapping filter; The region in the de-interferenced structural CT image corresponding to the non-zero label value region in the label mapping information is retained to obtain the minimum region of interest CT image.

17. The bone segmentation device for ankle joint CT images according to claim 13, characterized in that: The third bone information extraction unit is used to extract a bone structure CT image and a background label image for reference from the minimum region of interest CT image based on a third segmentation threshold interval, a second segmentation threshold and the number of the feet, including: According to the third segmentation threshold interval, the CT values ​​of the pixels in the minimum region of interest CT image are reset by using a threshold segmentation technique to obtain the reference bone structure CT image; According to the second segmentation threshold and the number of the feet, the CT values ​​of the pixels in the reference bone structure CT image are reset by using a threshold segmentation technique to obtain the background label image.

18. The bone segmentation device for ankle joint CT images according to claim 17, characterized in that: The method of resetting the CT values ​​of the pixels in the reference bone structure CT image by using a threshold segmentation technique according to the second segmentation threshold and the number of the feet to obtain the background label image includes: Determine the relationship between the pixel points in the currently traversed reference bone structure CT image and the second segmentation threshold; If the CT value of the pixel point is greater than the second segmentation threshold, the CT value of the pixel point is set to the first preset CT value; otherwise, the corresponding preset CT value is set according to the number of feet; Get the background label image.

19. The bone segmentation device for ankle joint CT images according to claim 11, characterized in that: The bone segmentation result generation module includes a noisy label image acquisition unit, which is used to acquire a noisy label image having the same parameters as the ankle joint CT image to be segmented according to the reference bone structure image and the seed image, including: Combining the reference bone structure CT image and the seed image, according to a region growing algorithm, taking the pixel point region of the seed image as the initial region, obtaining a noisy cropped image corresponding to different label values ​​for each bone structure; The noisy cropped image is resampled to obtain a noisy labeled image having the same parameters as the ankle joint CT image to be segmented.

20. The bone segmentation device for ankle joint CT images according to claim 11, characterized in that: The bone segmentation result generation module also includes an image post-processing unit, which is used to perform denoising and filling on the noisy label image, respectively, to obtain a label image containing label values ​​corresponding to different bone structures and a three-dimensional model file of each bone structure, including: Extracting multiple single-label images of each label in the noisy label image one by one, and smoothing and denoising each of the single-label images by using a median filter; Performing morphological closing operation on each of the single-label images processed by the median filter, and performing surface extraction to obtain a three-dimensional model surface of each label; Process the surface of each of the three-dimensional models by using a surface wrapping solidification algorithm to obtain a three-dimensional model file of each bone structure; The internal pixel points of each surface of the three-dimensional model after surface processing are filled with corresponding label values ​​to obtain a label image containing label values ​​corresponding to different bone structures.

Images