Computed tomography (CT) image reconstruction method, device and computer equipment

Abstract

The application relates to a CT image reconstruction method, device and computer equipment. The method comprises the following steps: acquiring at least one first CT image; processing the first CT image by using at least one contrast adjustment model to obtain at least one second CT image; extracting features of each second CT image to obtain a feature vector group; and performing decoding and reconstruction processing on the first CT image and the feature vector group to output a CT reconstructed image. The method can improve display contrast and further improve the reconstruction effect of the CT image.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a CT image reconstruction method, apparatus and computer equipment. Background Technology

[0002] With the development of CT (Computed Tomography) technology, CT imaging has been widely used in spinal imaging examinations due to its advantages of rapid imaging, low examination cost, and clear visualization of intervertebral disc accumulation and calcification. However, CT imaging often produces blurred images of conditions such as intervertebral disc degeneration, spinal cord deformation, dural sac compression, and nerve root compression.

[0003] Current CT imaging methods, or traditional methods, suffer from problems such as low display contrast. Summary of the Invention

[0004] Therefore, it is necessary to provide a CT image reconstruction method, apparatus, and computer equipment that can improve display contrast in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a CT image reconstruction method, the method comprising:

[0006] Acquire at least one first CT image;

[0007] The first CT image is processed using at least one contrast adjustment model to obtain at least one second CT image;

[0008] Features of each second CT image are extracted to obtain a feature vector set;

[0009] The first CT image and feature vector group are decoded and reconstructed to output the CT reconstructed image.

[0010] In one embodiment, the method further includes:

[0011] Acquire multiple raw CT images, as well as multiple sets of target contrast images adjusted from the raw CT images;

[0012] The original CT images and the target contrast images are input into a generative adversarial network for training, resulting in at least one contrast adjustment model.

[0013] In one embodiment, each set of target contrast images includes at least one image obtained by adjusting the contrast; the original CT images and the target contrast images are input into a generative adversarial network for training to obtain at least one contrast adjustment model, including:

[0014] The original CT images and the contrast images of each group of targets are input into the generative adversarial network for training, resulting in at least one contrast adjustment model.

[0015] In one embodiment, acquiring at least one first CT image includes:

[0016] Acquire the target CT image; the target CT image includes at least one intervertebral disc; perform intervertebral disc localization processing on the target CT image to obtain the center point of each intervertebral disc;

[0017] Based on the center point of each intervertebral disc and the preset rotation alignment rules, the target CT image is rotated and aligned sequentially to obtain the region of interest of each intervertebral disc.

[0018] By cropping the region of interest for each intervertebral disc, at least one first CT image of the intervertebral disc is obtained.

[0019] In one embodiment, the target CT image is processed to locate the intervertebral discs, obtaining the center points of each intervertebral disc, including:

[0020] Multi-target image segmentation is performed on the target CT image to obtain the vertebral body segmentation result; the vertebral body segmentation result includes CT images of multiple vertebrae and the vertebral body label corresponding to each vertebra.

[0021] The vertebral segmentation results are fitted to the minimum circumscribed cuboid of the vertebral body to obtain the center point of the vertebral body; the center point of the vertebral body is the centroid of the minimum circumscribed cuboid.

[0022] Based on the center point of each vertebra, the spine fitting curve is obtained;

[0023] The center point of each intervertebral disc is obtained based on the spinal fitting curve and the center point of each vertebral body.

[0024] In one embodiment, regions of interest for each intervertebral disc are cropped to obtain first CT images, including:

[0025] Obtain the lengths of each endplate adjacent to the intervertebral disc;

[0026] Generate a trimmed cuboid based on the maximum value among the lengths of each end plate;

[0027] The region of interest of the intervertebral disc was cropped using a cuboid to obtain the first CT image.

[0028] In one embodiment, features of each second CT image are extracted to obtain a feature vector set, including:

[0029] Each second CT image of the intervertebral disc is processed using a feature extractor to generate a feature vector set; the feature vector set includes multiple feature vectors for the second CT images; the number of column vectors in the feature vector set is the same as the number of contrast adjustment models.

[0030] In one embodiment, decoding and reconstructing the first CT image and the feature vector group to output a CT reconstructed image includes: inputting the soft tissue window CT image and the feature vector group into a variational autoencoder and outputting a CT reconstructed image; the soft tissue window CT image is an image obtained by adjusting the window width and window level of the first CT image based on preset parameters; the CT reconstructed image includes an image sequence of each intervertebral disc under the target contrast.

[0031] Secondly, this application provides a CT image reconstruction apparatus, the apparatus comprising:

[0032] The first CT image acquisition module is used to acquire at least one first CT image;

[0033] A local adjustment module is used to process a first CT image using at least one contrast adjustment model to obtain at least one second CT image.

[0034] The feature extraction module is used to extract features from each second CT image to obtain a set of feature vectors.

[0035] The fusion reconstruction module is used to decode and reconstruct the first CT image and the feature vector group, and output the CT reconstructed image.

[0036] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0037] The aforementioned CT image reconstruction method, apparatus, and computer equipment involve acquiring at least one first CT image; processing the first CT image using at least one contrast adjustment model to obtain at least one second CT image; extracting features from each second CT image to obtain a feature vector set; and performing decoding and reconstruction processing on the first CT image and the feature vector set to output a reconstructed CT image. This method can integrate the features of each second CT image generated by at least one contrast adjustment model, reducing the adjustment and image reading time for doctors on the initial CT image in traditional methods, improving the contrast of CT images on the target bone structure, and increasing the efficiency of obtaining the optimal contrast image for the target bone structure. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating a CT image reconstruction method in one embodiment;

[0039] Figure 2 This is a flowchart illustrating the CT image reconstruction steps in one embodiment;

[0040] Figure 3 This is a flowchart illustrating the step of acquiring at least one first CT image in one embodiment;

[0041] Figure 4 This is a flowchart illustrating the steps of performing intervertebral disc localization processing on a target CT image in one embodiment.

[0042] Figure 5(a) is a schematic diagram of the spine fitting curve in one embodiment;

[0043] Figure 5(b) is a schematic diagram of intervertebral disc rotation alignment in one embodiment;

[0044] Figure 5(c) is a schematic diagram of the region of interest for intervertebral disc trimming in one embodiment;

[0045] Figure 6 This is a flowchart illustrating the steps for trimming the regions of interest of each intervertebral disc in one embodiment.

[0046] Figure 7 This is a flowchart illustrating the steps for acquiring at least one first CT image in an example.

[0047] Figure 8 A flowchart of the overall process of CT image reconstruction in an example;

[0048] Figure 9 This is a structural block diagram of a CT image reconstruction device in one embodiment;

[0049] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] Spinal degenerative diseases encompass a wide range of conditions, among which intervertebral disc herniation is one of the most common. The diagnosis of the herniation level is crucial for confirming the diagnosis and selecting a treatment plan for spinal degenerative diseases. CT imaging, with its advantages of rapid imaging, low cost, and clear visualization of disc accumulation and calcification, has become an indispensable examination in the diagnosis of intervertebral disc herniation. Taking the diagnosis of disc herniation type in CT images as an example, the current clinical workflow involves radiologists observing each of the 23 intervertebral discs in each patient and recording the qualitative description of the disc type in the imaging report. The ever-increasing volume of spinal degenerative disease examinations in medical institutions is constantly increasing the workload of radiologists, and doctors under fatigue may miss or misdiagnose cases, delaying patient treatment. CT images often show blurred images of intervertebral disc degeneration, spinal cord deformation, dural sac compression, and nerve root compression. Clinicians often need to dynamically adjust the window width and level of CT images in real time when diagnosing disc bulging, protrusion, or extrusion. This relies heavily on the physician's experience and adjustment efficiency. Physicians need to observe the morphology and location of the intervertebral disc, dural sac, and spinal cord separately through adjustments, and finally make a comprehensive judgment to arrive at a diagnosis. This application proposes a CT image reconstruction method that improves the contrast of CT images without requiring manual adjustment by the physician, while simultaneously improving the efficiency of CT image contrast adjustment and enhancing the reconstruction effect.

[0052] In one embodiment, such as Figure 1 As shown, a CT image reconstruction method is provided, the method including:

[0053] Step 110: Acquire at least one first CT image;

[0054] Specifically, the first CT image can be a CT image of a target bony structure, such as an intervertebral disc, so that the contrast can be adjusted separately for each target bony structure in the subsequent process.

[0055] In some examples, the initial CT image can be acquired and segmented into a first CT image of multiple target bony structures by performing image segmentation processing on the initial CT image; the initial CT image can be a spinal CT image, and the first CT image can be a CT image targeting the intervertebral disc.

[0056] Step 120: Process the first CT image using at least one contrast adjustment model to obtain at least one second CT image;

[0057] Specifically, the contrast adjustment model can be an optimal contrast model for a local target, such as a contrast adjustment model for a single target tissue in the first CT image. By processing the first CT image using at least one contrast adjustment model, contrast adjustment can be performed on multiple target tissues in the first CT image separately. The second CT image is the image obtained after processing the first CT image using at least one contrast adjustment model, thus obtaining multiple locally optimal contrast images for the target bony structure. By adjusting the CT images of each target bony structure using at least one contrast model, multiple second CT images with contrast adjustment for different target tissues can be obtained. This reduces the adjustment and image reading time for doctors on the initial CT image of the target bony structure in traditional methods, facilitates the generation of the optimal imaging image for the target bony structure, and improves the efficiency of obtaining the optimal contrast image for the target bony structure.

[0058] In some examples, the first CT image may be a CT image targeting the intervertebral disc, and multiple second CT images may include one or more of the following: best-contrast images of the intervertebral disc, best-contrast images of the dural sac, and best-contrast images of the spinal cord.

[0059] Step 130: Extract the features of each second CT image to obtain a feature vector group;

[0060] Specifically, feature extraction can be performed on each second CT image separately to obtain a feature vector set including the feature vectors corresponding to each second CT image. By extracting features from each second CT image obtained after processing by at least one contrast adjustment model, it is easier to integrate the features of the second CT images for contrast adjustment targeting different target tissues, which is beneficial to improving the contrast of the final generated CT image targeting the target bony structure.

[0061] In some examples, feature extraction algorithms or models can be used to extract features from each second CT image separately, resulting in feature vectors for each second CT image. The resulting feature vectors are then combined into the aforementioned feature vector group.

[0062] Step 140: Decode and reconstruct the first CT image and feature vector group to output the CT reconstructed image.

[0063] Specifically, the first CT image can be decoded and reconstructed based on the feature vector group to obtain the CT reconstructed image. Since the feature vector group extracts the features of each second CT image, and each second CT image is processed using at least one contrast adjustment model, the CT reconstructed image incorporates the contrast adjustment features of each second CT image, thereby improving the contrast of the CT image for the target bony structure.

[0064] In some examples, the contrast of the first CT image can be adjusted using a traditional soft tissue window, and then simultaneously input into a depth generation model such as a variational autoencoder for decoding and reconstruction along with the feature vector group to obtain a CT reconstructed image; the CT reconstructed image can be a reconstructed image targeting the intervertebral disc.

[0065] In this embodiment, multiple first CT images are acquired; at least one contrast adjustment model is used to process the first CT images to obtain at least one second CT image; features of each second CT image are extracted to obtain a feature vector group; the first CT images and the feature vector group are decoded and reconstructed to output a CT reconstructed image. This method can integrate the features of each second CT image generated by at least one contrast adjustment model, reducing the adjustment time and image reading time for doctors on the target bone structure in the initial CT image in traditional methods, improving the contrast of CT images on the target bone structure, and improving the efficiency of obtaining the optimal contrast image for the target bone structure.

[0066] In one embodiment, such as Figure 2 As shown, the method also includes:

[0067] Step 210: Acquire multiple original CT images and multiple sets of target contrast images obtained by adjusting the original CT images;

[0068] Step 220: Input the original CT image and the target contrast image into the generative adversarial network for training to obtain at least one contrast adjustment model.

[0069] Specifically, multiple raw CT images can be used to adjust the contrast of different targets separately, resulting in multiple sets of adjusted target contrast images. The raw CT images and target contrast images can be input into a generative adversarial network (GAN) for training. The GAN can include a generator and a discriminator; the generator can include an encoder, which may include several convolutional layers; the discriminator can include a decoder, which may include several deconvolutional layers. For example, by inputting the raw CT images and each set of target contrast images into the GAN for training, at least one contrast adjustment model can be obtained. Multiple contrast adjustment models can be models that adjust contrast separately for different target tissues. By acquiring target contrast images for training and inputting them along with the raw CT images into the GAN for training, at least one contrast adjustment model can be obtained to improve the efficiency of contrast adjustment. Training the model based on the adjusted target contrast images is beneficial for accurately generating optimally visualized CT images of target bony structures, thereby improving the contrast of CT images for target bony structures.

[0070] In some examples, the constructed generative adversarial network (GAN) can include one or more of the following: Conditional Generative Adversarial Network (CGAN), Image Translation Pix2PixGAN, and Recurrent Generative Adversarial Network (CycleGAN). The discriminator can be a convolutional neural network framework; the loss function of the GAN can include a regularization function.

[0071] In one embodiment, each set of target contrast images includes at least one image obtained by adjusting the contrast; the original CT images and the target contrast images are input into a generative adversarial network for training to obtain at least one contrast adjustment model, including:

[0072] The original CT images and the contrast images of each group of targets are input into the generative adversarial network for training, resulting in at least one contrast adjustment model.

[0073] Specifically, multiple contrast adjustment models can be images obtained by adjusting the contrast for different target tissues. For example, the multiple contrast adjustment models can be contrast adjustment models for a first tissue, a second tissue, and a third tissue, or they can be contrast adjustment models for more or fewer target tissues. For example, the multiple contrast adjustment models can be a first contrast model for intervertebral disc contrast adjustment, a second contrast model for dural sac contrast adjustment, and a third contrast model for spinal cord contrast adjustment. A contrast adjustment dataset for the intervertebral disc can be constructed by acquiring target contrast images for the first tissue, the second tissue, and the third tissue, where the target contrast image for the first tissue can be an intervertebral disc target contrast image, the target contrast image for the second tissue can be a dural sac target contrast image, and the target contrast image for the third tissue can be a spinal cord target contrast image. For example, the optimal display image I1 for the intervertebral disc, the optimal display image I2 for the dural sac, and the optimal display image I3 for the spinal cord. The target contrast image for the first tissue can be based on empirical values. The optimal contrast image of the intervertebral disc obtained through adjustment can be generated using the following methods: First contrast model: This model can be obtained by training a generative adversarial network (GAN) with the original CT image and the target contrast image for the first tissue. Second contrast model: This model can be obtained by training a generative adversarial network (GAN) with the original CT image and the target contrast image for the second tissue. Third contrast model: This model can be obtained by training a generative adversarial network (GAN) with the original CT image and the target contrast image for the second tissue. Fourth contrast model: This model can be obtained by training a generative adversarial network (GAN) with the original CT image and the target contrast image for the third tissue. Fifth contrast model: This model can be obtained by training a generative adversarial network (GAN) with the original CT image and the target contrast image for the third tissue. Fifth contrast model: This model can be obtained by training a generative adversarial network (GAN) with the original CT image and the target contrast image for the third tissue. Fifth contrast model: This model can be obtained by training a generative adversarial network (GAN) with the target contrast image for the third tissue. Sixth contrast model: This model can be obtained by further processing the first CT image I using pre-trained generative adversarial models (first contrast model G1, second contrast model G2, and third contrast model G3). The resulting multiple second CT images of the intervertebral disc can include one or more of the optimal contrast images of the intervertebral disc (I1'), the optimal contrast image of the dura mater (I2'), and the optimal contrast image of the spinal cord (I3'). By generating a first contrast model, a second contrast model, and a third contrast model, the first CT image can be processed directly. This avoids clinicians having to perform real-time dynamic window width and window level adjustments on the target CT image to observe the morphology and location of the intervertebral disc, dural sac, and spinal cord separately. It reduces the adjustment time and image reading time for doctors in traditional methods and improves the efficiency of obtaining intervertebral disc images with the best contrast.

[0074] In some examples, the discriminator in the generative adversarial network (GAN) can be a convolutional neural network framework to distinguish the target contrast image In from the second CT image I' generated from the first CT image I after processing by the GAN. Regularization constraints can be added to the loss function of the GAN to make the second CT image I' as close as possible to the target contrast image In. The regularization function can be an L1 function or an L2 function. The target contrast images for the first tissue can be obtained by radiologists with senior experience (e.g., 7-10 years of clinical image interpretation experience) by adjusting the original CT images of all enrolled patients to achieve the optimal contrast enhancement state for the intervertebral disc, and the window width WW1 and window level WL1 under the optimal contrast enhancement state are recorded. The target contrast images for the second tissue can be obtained by radiologists with senior experience by adjusting the original CT images of all enrolled patients to achieve the optimal contrast enhancement state for the dural sac, and the window width WW2 and window level WL2 under the optimal contrast enhancement state are recorded. The target contrast images for the third tissue can be obtained by radiologists with senior experience by adjusting the original CT images of all enrolled patients to achieve the optimal contrast enhancement state for the spinal cord, and the window width WW3 and window level WL3 under the optimal contrast enhancement state are recorded.

[0075] In one embodiment, such as Figure 3 As shown, acquiring at least one first CT image includes:

[0076] Step 310: Acquire the target CT image; the target CT image includes at least one intervertebral disc;

[0077] Step 320: Perform intervertebral disc localization processing on the target CT image to obtain the center point of each intervertebral disc;

[0078] Step 330: Based on the center point of each intervertebral disc and the preset rotation alignment rules, the target CT image is rotated and aligned sequentially to obtain the region of interest of each intervertebral disc.

[0079] Step 340: Crop the region of interest of each intervertebral disc to obtain at least one first CT image of the intervertebral disc.

[0080] Specifically, target CT images can be acquired, which can be initial CT images of the spine. Intervertebral discs within the target CT images can be localized; for example, image processing methods can be used to mark the intervertebral discs in the target CT images to obtain the center point of each disc. The center point of each disc can be the three-dimensional coordinates of its center of gravity. A preset rotation alignment can be implemented using computer vision algorithms. Based on the center point of each disc, for example, using the center point as the rotation center, the target CT image can be rotated and aligned to obtain the region of interest (ROI) of the intervertebral disc, thus reducing the impact of different postures of the discs in three-dimensional space. Since the area of ​​each intervertebral disc in the target CT image occupies a relatively small proportion of the CT image, by accurately locating the relevant areas of the intervertebral discs in the target CT image and cropping the ROI of each disc to completely crop the relevant areas, the resulting first CT images facilitate targeted adjustment of the contrast of the intervertebral discs and related tissues. Through the above preprocessing, the target CT image can be segmented into multiple first CT images targeting the intervertebral discs, which facilitates subsequent contrast adjustment for each intervertebral disc.

[0081] In some examples, each of the cropped first CT images contains the complete image region of the corresponding intervertebral disc.

[0082] In one embodiment, such as Figure 4 As shown, the target CT image is processed to locate the intervertebral discs, obtaining the center points of each intervertebral disc, including:

[0083] Step 410: Perform multi-target image segmentation on the target CT image to obtain the vertebral body segmentation result; the vertebral body segmentation result includes CT images of multiple vertebrae and the vertebral body label corresponding to each vertebra.

[0084] Step 420: Fit the vertebral segmentation results to the minimum circumscribed cuboid of the vertebral body to obtain the center point of the vertebral body; the center point of the vertebral body is the centroid of the minimum circumscribed cuboid.

[0085] Step 430: Based on the center point of each vertebra, obtain the spine fitting curve;

[0086] Step 440: Based on the spinal fitting curve and the center point of each vertebra, obtain the center point of each intervertebral disc.

[0087] Specifically, multi-object image segmentation can label each pixel in an image as one of several pre-determined object categories, simultaneously performing the identification and segmentation of multiple targets. Multi-object image segmentation of target CT images can yield vertebral body segmentation results. These results include CT images of multiple vertebrae. Multi-label image classification of the target CT images ensures that the vertebral body segmentation results include the vertebral body label corresponding to each vertebra. As shown in Figure 5(a), for each vertebral body CT image in the segmentation results, a minimum circumscribed cuboid can be fitted for each vertebra, and then the centroid of the minimum circumscribed cuboid can be used to determine the... The center points of the vertebral bodies are fitted to obtain a spinal fitting curve. For example, the center points of each vertebra can form a point set, which can be fitted using interpolation methods to obtain a smooth fitting curve C for the spine. Multiple intervertebral disc center points can be determined based on the spinal fitting curve and the center points of each vertebra. For example, the midpoint between the center points of adjacent vertebrae can be selected. If the midpoint lies on the spinal fitting curve, it can be determined as the center point of the intervertebral disc. If the midpoint lies outside the spinal fitting curve, the point on the curve with the smallest distance from the midpoint can be determined as the center point of the intervertebral disc. Through these steps, the center point of the intervertebral disc can be determined quickly and accurately, improving the efficiency and accuracy of intervertebral disc localization.

[0088] In some examples, a trained deep learning segmentation model can be used to segment the target CT image to obtain vertebral body segmentation results. Different label values ​​can represent different vertebrae in the segmentation results. Further optimization of the vertebral body segmentation results can be achieved using graph cut algorithms or deep learning algorithms to avoid undersegmentation or oversegmentation leading to inaccurate vertebral body segmentation. Vertebral detection and localization methods can be used to obtain the three-dimensional coordinates of the vertebral body center points. For example, a deep learning object detection model or a deep learning point detection model can be used to obtain the center points of each vertebra. Interpolation methods can include at least one of B-spline interpolation and polynomial interpolation. A point set P[P1, P2, ..., P5] can be formed by combining the center points of the 23 vertebrae (C2-C7, T1-T12, L1-L5) from C2 to L5. 23 The point set P is fitted using an interpolation method to form the smooth fitted curve C described above. For the point determined to be the center of the intervertebral disc, P0 can be selected sequentially from the point set P. n and P n+1 P, the midpoint of (n = 1 to 22) C If the midpoint P C If the midpoint P falls exactly on the fitted curve C, then... C 'Consider P as the center point of the intervertebral disc o-min Otherwise, select P from the fitted curve C. Cn To PCn+1 On the truncated segment, with P C The point with the smallest absolute distance is taken as the center point P of the intervertebral disc. o-min .

[0089] In some examples, as shown in Figure 5(b), the target CT image can be rotated and aligned based on the fitted curve C. The fitted curve C at the center point P of the intervertebral disc can be obtained. o-min If the angle between the tangent vector V' and the direction vector V of the fitted curve C is θ, then the original CT image is rotated around the orthogonal vector obtained from the tangent vector V' and the direction vector V, with the intervertebral disc center point P as the axis of rotation. o-min Rotate and align the center of rotation, for example, by rotating by an angle θ, to obtain the region of interest of the intervertebral disc.

[0090] In one embodiment, such as Figure 6 As shown, regions of interest for each intervertebral disc are cropped to obtain the first CT images, including:

[0091] Step 610: Obtain the lengths of each endplate adjacent to the intervertebral disc;

[0092] Step 620: Generate a trimmed cuboid based on the maximum value among the lengths of each final plate;

[0093] Step 630: The region of interest of the intervertebral disc is cropped using a cuboid to obtain the first CT image.

[0094] Specifically, the endplates adjacent to the intervertebral disc constitute the superior and inferior bony boundaries of the disc, including the inferior endplate of the vertebral body above the disc and the superior endplate of the vertebral body below the disc. The maximum length among the endplates adjacent to the disc can be selected as a reference to generate a trimmed cuboid. For example, as shown in Figure 5(c), the length and width of the trimmed cuboid can be set to twice the maximum value, and the height can be set to the maximum value. This trimmed cuboid is then used to trim the region of interest (ROI) of the intervertebral disc, yielding the first CT image. The length, width, and height of the trimmed cuboid can also be set according to actual needs to include the complete intervertebral disc image region. These steps ensure that the ROI contains similar intervertebral disc information, reducing the influence of other image information.

[0095] In some examples, the lengths of the inferior endplates of the vertebral bodies above the intervertebral disc and the superior endplates of the vertebral bodies below the intervertebral disc can be determined based on CT images of the intervertebral disc in the midsagittal plane using computer vision and other related image algorithms.

[0096] In one embodiment, features of each second CT image are extracted to obtain a feature vector set, including:

[0097] Each second CT image of the intervertebral disc is processed using a feature extractor to generate a feature vector set; the feature vector set includes multiple feature vectors for the second CT images; the number of column vectors in the feature vector set is the same as the number of contrast adjustment models.

[0098] Specifically, a feature extractor can be used to process each second CT image of the intervertebral disc separately, generating a feature vector set. Each feature vector in the feature vector set corresponds one-to-one with each second CT image. The number of column vectors in the feature vector set can be the same as the number of feature vectors in the feature vector set, which is the same as the number of contrast adjustment models. In the above steps, by improving the comprehensiveness of the feature description of each second CT image by the feature vector set, the features of contrast adjustment are accurately extracted, thereby improving the final contrast of the intervertebral disc CT image and improving the reconstruction effect of the CT image.

[0099] In some examples, a deep feature extractor can be used to process the optimal contrast images of the intervertebral disc (I1'), the dural sac (I2'), and the spinal cord (I3') separately, abstracting a 3×N dimensional feature vector set F. The feature vector set F can include three feature vectors, for example, a 1×N dimensional feature vector F1, a 1×N dimensional feature vector F2, and a 1×N dimensional feature vector F3. The deep feature extractor includes multiple cascaded convolutional layers and fully connected layers.

[0100] In one embodiment, decoding and reconstructing the first CT image and the feature vector group to output a CT reconstructed image includes: inputting the soft tissue window CT image and the feature vector group into a variational autoencoder and outputting a CT reconstructed image; the soft tissue window CT image is an image obtained by adjusting the window width and window level of the first CT image based on preset parameters; the CT reconstructed image includes an image sequence of each intervertebral disc under the target contrast.

[0101] Specifically, the soft tissue window CT image Is can be obtained by adjusting the window width and window level of the first CT image I of the intervertebral disc. The soft tissue window CT image Is can be paired with a feature vector group F and input into a variational autoencoder (VAE). For example, the soft tissue window CT image Is can be paired with feature vectors F1, F2, and F3 in the feature vector group F as input to the VAE to obtain the reconstructed image Io of the intervertebral disc. The reconstructed image Io can include the image sequence of each intervertebral disc in the spine at a target contrast (e.g., optimal contrast). For example, the reconstructed image Io can be a sequence composed of each intervertebral disc Ion (n = 1 to 23) on a per-object basis, and can be displayed as a sub-view in the radiologist's viewing system. The above steps can fuse the feature vectors, improve the contrast adjustment effect of the soft tissue window CT image, and generate a reconstructed image of the intervertebral disc that has good imaging effect on the intervertebral disc, dural sac, and spinal cord.

[0102] In some examples, adjusting the preset parameters for the window width and window level of the first CT image may include: a soft tissue window width (WW) of 300–500 HU and a soft tissue window level (WL) of 40–60 HU. The variational autoencoder may include an encoder and a decoder; random noise (e.g., Gaussian noise) may be added to the encoder to enhance the robustness of the generated image, and the KL function may be selected as the loss function of the variational autoencoder. The reconstructed intervertebral disc image (Io) can be displayed in a multiplanar reformation (MPR) manner, facilitating simultaneous viewing of the axial, coronal, and sagittal planes of the reconstructed cervical spine CT image.

[0103] In some examples, such as Figure 7The diagram illustrates the process of acquiring at least one first CT image. First, multi-target segmentation of the spinal vertebrae is performed to obtain vertebral body segmentation results. These results are then optimized, for example, using graph cut algorithms or deep learning algorithms, to avoid under-segmentation or over-segmentation that could lead to inaccurate segmentation of the vertebrae, resulting in multi-target vertebral body segmentation results. Next, a minimum circumscribed cuboid is fitted to the vertebral body segmentation results to obtain the coordinates of the vertebral body's center point. Based on the vertebral body center point coordinates and the multi-target vertebral body segmentation results, the intervertebral discs are located to obtain their center points. The intervertebral discs in the CT image are then rotated and aligned; for example, each intervertebral disc in the CT image is rotated and aligned using its center point as the rotation center. Finally, the intervertebral discs in the CT image are cropped to obtain CT images of each disc. This preprocessing of the target CT image allows for accurate segmentation into multiple intervertebral disc CT images, facilitating subsequent contrast adjustment for each disc.

[0104] In some examples, such as Figure 8 The diagram shown illustrates the overall flowchart of CT image reconstruction. The intervertebral disc in the CT image can be sequentially located, rotated, and cropped to obtain the intervertebral disc CT image I. The soft tissue window of the intervertebral disc CT image I can be adjusted to obtain the intervertebral disc soft tissue window CT image Is. The optimal display images I1, I2, and I3 of the intervertebral disc, obtained by radiologists through adjustments to the intervertebral disc CT image, can be acquired. Based on these optimal display images, individual optimal contrast generation models can be constructed, including optimal imaging models for the intervertebral disc, dural sac, and spinal cord. Inputting the intervertebral disc CT image I into each of these constructed individual optimal contrast generation models yields corresponding individual optimal contrast images, including optimal contrast images I1', I2', and I3' for the intervertebral disc, dural sac, and spinal cord. Fusing these individual optimal contrast images and inputting them along with the intervertebral disc soft tissue window CT image Is into the constructed VAE generation model results in images before and after contrast adjustment of the intervertebral disc CT image. Using the contrast-adjusted intervertebral disc CT image for image display can improve the intervertebral disc display contrast. In practical applications, the above overall flowchart only requires one model training. The trained model can be used directly and can also improve the efficiency of contrast adjustment, thereby improving the efficiency of CT image reconstruction.

[0105] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0106] Based on the same inventive concept, this application also provides a CT image reconstruction apparatus for implementing the CT image reconstruction method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more CT image reconstruction apparatus embodiments provided below can be found in the limitations of the CT image reconstruction method described above, and will not be repeated here.

[0107] In one embodiment, a CT image reconstruction apparatus is provided, such as Figure 9 As shown, the device includes:

[0108] The first CT image acquisition module 910 is used to acquire at least one first CT image;

[0109] The local adjustment module 920 is used to process the first CT image using at least one contrast adjustment model to obtain at least one second CT image.

[0110] Feature extraction module 930 is used to extract features from each of the second CT images to obtain a feature vector group;

[0111] The fusion reconstruction module 940 is used to decode and reconstruct the first CT image and the feature vector group, and output the CT reconstructed image.

[0112] In one embodiment, the device further includes:

[0113] The image acquisition module is used to acquire multiple raw CT images, as well as multiple sets of target contrast images adjusted from the raw CT images.

[0114] The model training module is used to input the original CT images and the target contrast images into the generative adversarial network for training, so as to obtain at least one contrast adjustment model.

[0115] In one embodiment, the model training module is further configured to train a generative adversarial network by inputting the original CT images and each group of target contrast images, thereby obtaining at least one contrast adjustment model. In one embodiment, the first CT image acquisition module 910 includes:

[0116] An image acquisition unit is used to acquire a target CT image; the target CT image includes at least one intervertebral disc.

[0117] The positioning processing unit is used to perform intervertebral disc localization processing on the target CT image to obtain the center point of each intervertebral disc.

[0118] The rotation alignment unit is used to sequentially rotate and align the target CT image based on the center point of each intervertebral disc and the preset rotation alignment rules to obtain the region of interest of each intervertebral disc.

[0119] The target clipping unit is used to clip the region of interest of each intervertebral disc to obtain at least one first CT image of the intervertebral disc.

[0120] In one embodiment, the positioning processing unit includes:

[0121] The cone segmentation subunit is used to perform multi-target image segmentation on the target CT image to obtain the vertebral body segmentation result; the vertebral body segmentation result includes CT images of multiple vertebrae and the vertebral body label corresponding to each vertebra.

[0122] The first center determines the sub-unit, which is used to fit the smallest circumscribed cuboid of the vertebral body to the vertebral body segmentation result to obtain the center point of the vertebral body; the center point of the vertebral body is the centroid of the smallest circumscribed cuboid.

[0123] The curve fitting subunit is used to obtain the spine fitting curve based on the center point of each vertebra;

[0124] The second center determines the sub-unit, which is used to obtain the center point of each intervertebral disc based on the spinal fitting curve and the center point of each vertebral body.

[0125] In one embodiment, the target cropping unit includes:

[0126] The length acquisition subunit is used to acquire the length of each endplate adjacent to the intervertebral disc;

[0127] The trimmed cuboid generation sub-unit is used to generate a trimmed cuboid based on the maximum value among the lengths of each final plate;

[0128] The target clipping subunit is used to clip the region of interest of the intervertebral disc using a clipping cuboid to obtain the first CT image.

[0129] In one embodiment, the feature extraction module 930 is further configured to process each of the second CT images of the intervertebral disc using a feature extractor to generate a feature vector group; the feature vector group includes multiple feature vectors for the second CT images; the number of column vectors in the feature vector group is the same as the number of contrast adjustment models.

[0130] In one embodiment, the fusion reconstruction module 940 is further configured to: input the soft tissue window CT image and the feature vector group into the variational autoencoder and output the CT reconstruction image; the soft tissue window CT image is an image obtained by adjusting the window width and window level of the first CT image based on preset parameters; the CT reconstruction image includes the image sequence of each intervertebral disc under the target contrast.

[0131] Each module in the aforementioned CT image reconstruction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0132] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0133] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a CT image reconstruction method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0134] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0135] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0136] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0137] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0138] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0139] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A CT image reconstruction method, characterized in that, The method includes: Acquire at least one first CT image; the first CT image includes a CT image of the target bony structure; The first CT image is processed using at least one contrast adjustment model to obtain at least one second CT image; the contrast adjustment model is obtained based on generative adversarial network training; wherein, when multiple contrast adjustment models are used to process the first CT image, the multiple contrast adjustment models are used to adjust the contrast for different target tissues respectively; Extract the features of each of the second CT images to obtain a feature vector group; The soft tissue window CT image and the feature vector group are paired and then input into the variational autoencoder to output the CT reconstructed image; the soft tissue window CT image is an image obtained by adjusting the window width and window level of the first CT image based on preset parameters; the CT reconstructed image includes the image sequence of each intervertebral disc under the target contrast.

2. The method according to claim 1, characterized in that, The method further includes: Acquire multiple raw CT images, and multiple sets of target contrast images adjusted based on the raw CT images; The original CT image and the target contrast image are input into a generative adversarial network for training to obtain at least one contrast adjustment model.

3. The method according to claim 2, characterized in that, Each group of target contrast images includes at least one image obtained by adjusting contrast; the step of inputting the original CT image and the target contrast image into a generative adversarial network for training to obtain at least one contrast adjustment model includes: The original CT images and the target contrast images of each group are input into the generative adversarial network for training to obtain at least one contrast adjustment model.

4. The method according to claim 1, characterized in that, The acquisition of at least one first CT image includes: Acquire a target CT image; the target CT image includes at least one intervertebral disc; The target CT images are processed to locate the intervertebral discs, and the center points of each intervertebral disc are obtained. Based on the center point of each intervertebral disc and the preset rotation and alignment rules, the target CT image is rotated and aligned sequentially to obtain the region of interest of each intervertebral disc. By cropping the region of interest of each intervertebral disc, at least one first CT image of the intervertebral disc is obtained.

5. The method according to claim 4, characterized in that, The step of performing intervertebral disc localization processing on the target CT image to obtain the center point of each intervertebral disc includes: Multi-target image segmentation is performed on the target CT image to obtain vertebral body segmentation results; the vertebral body segmentation results include CT images of multiple vertebrae and vertebral body labels corresponding to each vertebra. The vertebral body segmentation results are fitted with the minimum circumscribed cuboid of the vertebral body to obtain the center point of the vertebral body; the center point of the vertebral body is the centroid of the minimum circumscribed cuboid. Based on the center point of each vertebra, a spinal fitting curve is obtained; The center point of each intervertebral disc is obtained based on the spinal fitting curve and the center point of each vertebral body.

6. The method according to claim 4, characterized in that, The process of cropping the region of interest for each intervertebral disc to obtain each of the first CT images includes: Obtain the lengths of each endplate adjacent to the intervertebral disc; Generate a trimmed cuboid based on the maximum value among the lengths of each of the aforementioned end plates; The region of interest of the intervertebral disc is cropped using the aforementioned cuboid to obtain the first CT image.

7. The method according to claim 1, characterized in that, The extraction of features from each of the second CT images to obtain a feature vector group includes: Each of the second CT images of the intervertebral disc is processed using a feature extractor to generate the feature vector group; the feature vector group includes multiple feature vectors for the second CT images; the number of column vectors in the feature vector group is the same as the number of contrast adjustment models.

8. A CT image reconstruction device, characterized in that, The device includes: A first CT image acquisition module is used to acquire at least one first CT image; the first CT image includes a CT image targeting a bony structure. A local adjustment module is used to process the first CT image using at least one contrast adjustment model to obtain at least one second CT image; the contrast adjustment model is obtained based on generative adversarial network training; wherein, when multiple contrast adjustment models are used to process the first CT image, the multiple contrast adjustment models are used to adjust the contrast for different target tissues respectively; The feature extraction module is used to extract features from each of the second CT images to obtain a feature vector group; The fusion reconstruction module is used to decode and reconstruct the first CT image and the feature vector group, and output the CT reconstruction image. The fusion reconstruction module is further used to input the soft tissue window CT image and the feature vector group into the variational autoencoder and output the CT reconstruction image; the soft tissue window CT image is an image obtained by adjusting the window width and window level of the first CT image based on preset parameters; the CT reconstruction image includes the image sequence of each intervertebral disc under the target contrast.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images