Method, system and apparatus for oral cavity CBCT ultralow-dose imaging

Abstract

The present invention relates to the technical field of medical imaging. Disclosed are a method, system and apparatus for oral cavity CBCT ultralow-dose imaging. The method comprises: collection of high-dose oral cavity CBCT image data, data splitting processing, data three-dimensional reconstruction processing, model training, image data inputting, image enhancement, and image data outputting; and the specific process involves: performing three-dimensional reconstruction processing on oral cavity CBCT two-dimensional projection data in an ultralow-dose mode of a patient, so as to obtain ultralow-dose oral cavity CBCT three-dimensional reconstruction image data, inputting same into an oral cavity CBCT ultralow-dose imaging image enhancement network model, and performing outputting to obtain high-quality oral cavity CBCT image data for a subsequent clinical diagnosis and treatment process. The present invention can significantly reduce a radiation dose for imaging and accelerate an imaging speed while ensuring the imaging quality, thereby improving the safety of oral cavity CBCT imaging, and having broad clinical application prospects.

Description

Method, system and device for oral CBCT ultra-low-dose imaging Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a method, system and device for oral CBCT ultra-low-dose imaging. Background Art

[0002] Oral CBCT (Cone Beam Computed Tomography) is a commonly used dental imaging technique, widely used in the diagnosis and treatment of oral diseases. It uses a rotating cone-beam X-ray to scan the patient's oral area, obtaining a two-dimensional projection image. It then uses three-dimensional reconstruction technology to transform the two-dimensional projection image into a high-resolution three-dimensional image of the oral structure. Compared to traditional two-dimensional oral X-rays, oral CBCT provides more accurate and comprehensive information about oral structures and lesions, assisting doctors in making accurate diagnoses and surgical planning. It plays a vital role in the diagnosis of oral diseases, surgical planning, and implant surgery navigation.

[0003] However, existing oral CBCT imaging technology has a major problem: to obtain high-quality reconstructed images, a long X-ray exposure time and a high X-ray radiation dose are required. This poses a great risk to the patient's health, especially for some special populations, such as children and pregnant women. In particular, in situations where multiple repeated imaging is required, such as in the planning and intraoperative navigation of oral implant surgery, the cumulative effect of radiation may cause significant damage to the patient. High-dose radiation has a certain damaging effect on human tissues and organs. Children, young people and pregnant women are more sensitive to high-dose radiation. Long-term exposure to high-dose radiation may cause DNA damage and increase the risk of cancer. In addition, high-dose radiation may also cause damage and dysfunction to organs such as the larynx, thyroid and salivary glands.

[0004] To reduce radiation dose and potential risks to patients, researchers have been working to develop ultra-low-dose oral CBCT imaging methods to lower radiation dose while maintaining image quality. The research and development of these methods has important clinical significance for improving the safety, accuracy, and reliability of oral medical imaging.

[0005] Currently, some of the more common low-dose imaging methods for oral CBCT include improvements in reconstruction algorithms, image processing technology, and the use of more advanced imaging sensors. However, these methods have some limitations to a certain extent. Reconstruction algorithm improvements mainly improve the resolution and quality of images by improving back projection and filtering algorithms, thereby reducing the need for repeated scans and increased doses. Image processing technology mainly improves image quality through image denoising, enhancement, and model reconstruction, so that clear oral CBCT images can still be obtained under low-dose conditions. Another common low-dose imaging method is projection parameter optimization, which mainly refers to reducing radiation dose by optimizing scanning parameters, such as reducing the tube current and / or reducing the number of projection samples (i.e., sparse sampling).

[0006] However, these methods often introduce additional computational overhead and have limited effectiveness; the use of advanced sensors requires high costs and complex technical equipment; and projection parameter optimization, while reducing dose, often leads to loss of image detail and increased noise, resulting in reduced image quality and severe artifacts, which in turn affect the accuracy and reliability of clinical diagnosis. Therefore, developing a method that can significantly reduce oral CBCT dose while maintaining image quality is a current research hotspot and challenge.

[0007] In recent years, deep learning has made significant progress in the field of medical imaging. By training neural networks, deep learning can learn complex feature representations from large amounts of data, thereby improving image quality and reducing noise. In particular, the emergence of generative adversarial networks (GANs) has provided a new approach for image reconstruction and enhancement.

[0008] Combining deep learning with sparse sampling techniques is an effective approach in the study of ultra-low-dose oral CBCT imaging. Sparse sampling is a method for reducing the amount of data collected, significantly reducing the radiation dose of oral CBCT. However, sparse sampling results in low-quality data, with significant loss of image clarity and detail. Therefore, utilizing deep learning to recover high-quality images from sparsely sampled data has become an important area of ​​research in the field of ultra-low-dose oral CBCT imaging.

[0009] Therefore, the present invention proposes a method, system, and apparatus for ultra-low-dose oral CBCT imaging. This method, based on deep learning technology, utilizes a proprietary residual dense generative adversarial network (RDN-GAN) model to enhance the quality of sparsely sampled oral CBCT images. By splitting and reconstructing high-dose oral CBCT data collected from patients to generate paired image datasets, the RDN-GAN model is trained. This model can recover high-quality oral CBCT images from sparsely sampled data, significantly reducing radiation dose to patients and providing a more reliable and safer imaging technology for oral medicine.

[0010] Summary of the Invention

[0011] The purpose of the present invention is to provide a method, system and device for oral CBCT ultra-low-dose imaging, so as to solve the problem that existing oral CBCT imaging technology cannot reduce the radiation dose of CBCT imaging while maintaining imaging quality.

[0012] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0013] The present invention provides an oral CBCT ultra-low-dose imaging method, comprising:

[0014] Image data collection: collecting oral CBCT two-dimensional projection data of patients in high-dose mode;

[0015] Image data splitting processing: splitting the collected CBCT two-dimensional projection data of the patient in high-dose mode into paired ultra-sparse sampling projection data and full sampling projection data;

[0016] 3D reconstruction of image data: performing 3D reconstruction of the paired ultra-sparse sampling projection data and the fully sampled projection data to obtain paired ultra-sparse sampling oral CBCT reconstruction data and a fully sampled oral CBCT reconstruction data set;

[0017] Model training, using the paired ultra-sparse sampling oral CBCT reconstruction data and the fully sampled oral CBCT reconstruction data set as training samples to train a residual dense generative adversarial deep learning neural network model to obtain an oral CBCT ultra-low dose imaging image enhancement network model;

[0018] Image data input, performing three-dimensional image data reconstruction processing on the patient's oral CBCT two-dimensional projection data in ultra-low-dose mode to obtain ultra-low-dose oral CBCT reconstructed data, and inputting the data into the oral CBCT ultra-low-dose imaging image enhancement network model;

[0019] Image enhancement: the oral CBCT ultra-low-dose imaging image enhancement network model performs image enhancement processing on the ultra-low-dose oral CBCT reconstruction data to obtain a high-quality oral CBCT image of the patient;

[0020] Image data output: display and output the high-quality oral CBCT images of the patient.

[0021] Furthermore, the number of projections of the ultra-sparsely sampled projection data ranges from 20 to 60 projections, and the number of projections of the fully sampled projection data ranges from 360 to 720 projections;

[0022] The algorithms used in the three-dimensional reconstruction processing of the image data include a CBCT iterative reconstruction algorithm and an FDK three-dimensional reconstruction algorithm.

[0023] Furthermore, the residual dense generative adversarial deep learning neural network model includes a residual dense network and an adversarial network connected in sequence, and the loss function of the residual dense generative adversarial deep learning neural network model is:

[0024] Among them, N is the total number of training input samples, θ G Represents the weights and biases of the residual dense generative adversarial deep learning neural network, n is the index of each training sample, l SR is the perceptual loss function, represents the output of the residual dense network, The ultra-sparse sampling oral CBCT reconstruction data input sample of the training set, is the paired full-sampled oral CBCT reconstruction data sample, is the optimal solution of the trained network model.

[0025] Furthermore, the residual dense network includes a convolutional layer, a residual dense block, a cascade network, an upsampling layer, and a deconvolution layer, and the residual dense block is composed of a convolutional layer, an activation function, and a cascade layer;

[0026] The adversarial network consists of a convolutional layer, an activation function, a normalization layer, and a density network. The input and output of each density network are: d,c =σ(W d,c [F d-1 ,F d,1 ,…,F d,c-1 ])

[0027] Among them, F d,c is the output of the dth layer density network and the cth layer convolution network, F d-1 and F d are the input and output of the d-th layer density network, σ is the RELU activation function, Wd,c is the weight of each convolutional layer of the density network, [F d-1 ,F d,1 ,…,F d,c-1 ] is the cascade feature map generated by the d-1th layer density network; F d,1 is the output of the first convolutional network in the d-th density network, F d,c-1 is the output of the c-1th convolutional network in the dth density network.

[0028] The RELU activation function is: f(x)=max(0,x)

[0029] Among them, the RELU activation function is a linear function. Compared with the general Sigmoid or Tanh activation function, the RELU activation function does not have the problem of derivative vanishing or derivative exploding during training, which can make the entire training process more stable. In addition, the calculation of the RELU activation function is simpler and does not require floating-point operations, which greatly reduces the processing time.

[0030] The cascade network links the features of all density networks and adaptively controls the output information through a 1×1 convolutional network. Finally, the output of the entire local adversarial density network is obtained through a residual learning network.

[0031] Furthermore, in order to enhance the network capability, features are constructed before the RELU activation function and the perceptual loss function l is introduced. SR , and its calculation formula is:

[0032] in is the content loss function, To counter the loss function,

[0033] here Represents the reconstructed image, is the probability of a real picture, represents the output of the residual dense network, I LR represents the input image of the adversarial network, θ G Represents the weights and biases of the residual dense network, N is the total number of training inputs, n is the index of each training sample, and is calculated using the MSE model and PSNR model Value: PSNR=20*log 10 (MAX I )-10*log 10 (MSE)

[0034] Among them, MAX Iis the maximum value of the pixel. Since the data will be normalized during the training process, the maximum value is 1. m, n represent the horizontal and vertical pixel numbers of the input image resolution, respectively. i, j are the horizontal and vertical index numbers corresponding to each pixel. I, K represent the image after image enhancement and the fully sampled image, respectively.

[0035] Furthermore, the paired ultra-sparse sampling oral CBCT reconstruction dataset and the fully sampled oral CBCT reconstruction dataset are used as training samples to train a residual dense generative adversarial deep learning neural network model to obtain an oral CBCT ultra-low dose imaging image enhancement neural network model, including:

[0036] Performing data expansion and normalization processing on the paired ultra-sparse sampling oral CBCT reconstruction dataset and the fully sampled oral CBCT reconstruction dataset to obtain an image set; then dividing the normalized image set into a training set, a validation set, and a test set;

[0037] The training set is used to train the RDN-GAN deep learning network model, the validation set is used to verify the network model's effectiveness, and the test set is used to further test the network model's effectiveness. During the RDN-GAN network model training process, the Peak Signal to Noise Ratio (PSNR) is used as the loss function. The enhanced Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index (SSIM) are calculated on the validation and test sets to evaluate the network model's effectiveness.

[0038] The verification model is verified and tested using the verification set and the test set. If the verification result does not meet the prediction probability threshold, the generative network and the adversarial network are retrained. If the verification result meets the prediction probability threshold, further testing is performed using the test set. If the test result meets the prediction probability threshold, the training of the oral CBCT ultra-low dose imaging image enhancement network model is completed. If the test result does not meet the prediction probability threshold, the generative network and the adversarial network are retrained.

[0039] The present invention also provides a system for oral CBCT ultra-low-dose imaging, comprising: a projection acquisition module, a three-dimensional reconstruction module, an image enhancement module, and an image display and output module connected in sequence; the projection acquisition module is used to collect oral CBCT two-dimensional projection data of a patient in an ultra-low-dose mode; the three-dimensional reconstruction module is used to perform three-dimensional reconstruction on the oral CBCT two-dimensional projection data in the ultra-low-dose mode; the image enhancement module is used to input the oral CBCT two-dimensional projection data in the ultra-low-dose mode into an oral CBCT ultra-low-dose imaging image enhancement network model to obtain an image-enhanced oral CBCT image; and the image display and output module is used to display the image-enhanced oral CBCT image and output it in a Dicom file format.

[0040] The present invention also provides a device for oral CBCT ultra-low-dose imaging, comprising: an oral CBCT imaging component, an image reconstruction component, an image display component, and an image enhancement component; the oral CBCT imaging component is communicatively connected to the image reconstruction component, and the image reconstruction component is communicatively connected to the image display component and the image enhancement component respectively;

[0041] The image reconstruction component is used to perform three-dimensional reconstruction of the data acquired by the oral CBCT imaging component; the image enhancement component stores an oral CBCT ultra-low-dose imaging image enhancement network model and performs image enhancement processing on the three-dimensional data reconstructed by the image reconstruction component; the image display component is used to display the image after image enhancement processing.

[0042] Furthermore, the oral CBCT imaging assembly includes an X-ray tube, a flat-panel detector, a bracket and a motion platform. The X-ray tube is arranged on the motion platform and is located on one side of the patient. The motion platform is arranged on the bracket. The flat-panel detector is arranged on the other side of the patient and matches the X-ray tube.

[0043] Furthermore, the image enhancement component includes a memory, a processor and a network interface, and the memory, processor and network interface are electrically connected; the memory stores an oral CBCT ultra-low-dose imaging image enhancement network model.

[0044] Furthermore, the memory includes: phase change memory, static random access memory, dynamic random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory, compact disc read-only memory or digital versatile disc.

[0045] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0046] The present invention discloses a method, system, and device for ultra-low-dose oral CBCT imaging. The imaging method includes image data acquisition, image data segmentation and processing, three-dimensional reconstruction of image data, model training, image data input, image enhancement, and image data output. The system comprises a sequentially connected projection acquisition module, a three-dimensional reconstruction module, an image enhancement module, and an image display and output module. The device includes an oral CBCT imaging component, an image reconstruction component, an image display component, and an image enhancement component.

[0047] 1) More high-quality patient clinical oral CBCT image data was used to train the algorithm model. In this application, large-scale clinical data (over 1,500 cases, of which over 200 were high-definition data) was used for model training. This dataset is also one of the largest in the world and can effectively improve the training effect of the model;

[0048] 2) An iterative reconstruction algorithm is used for 3D reconstruction of 2D projection data. This algorithm achieves better reconstruction clarity than the traditional FDK reconstruction algorithm. In the case of sparse sampling, the iterative reconstruction algorithm has a higher tolerance for sparse sampling, enabling high-definition imaging with a lower number of samples. However, the main drawbacks of the iterative reconstruction algorithm are high computing power requirements and long reconstruction time, especially in HD mode. In the present invention, the computing power requirements and time consumption of iterative reconstruction are greatly reduced due to the significant reduction in the number of samples, thus achieving a balance between reconstruction time and image quality.

[0049] 3) This invention utilizes the latest improved deep learning algorithm model based on the residual dense generative adversarial network architecture. This algorithm has demonstrated superior performance to all existing deep learning algorithms in the field of traditional computer vision. This invention improves and refines this algorithm and applies it to the field of oral CBCT imaging enhancement, which can more effectively improve image quality and enhance image details. Compared with previous CNN deep learning image enhancement algorithms, RDN-GAN can better maintain the spatial resolution of imaging and can obtain high-quality oral CBCT images under extremely sparse sampling conditions.

[0050] 4) The oral CBCT ultra-low-dose imaging method of the present invention can significantly reduce the radiation dose of patients undergoing oral CBCT examinations, with the maximum reduction being one twentieth to one thirtieth of the existing dose. It can also reduce imaging time and motion-related artifacts while maintaining very high imaging clarity, thereby improving the safety of patients, especially children or those who require frequent imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The present invention will be further described below with reference to the accompanying drawings.

[0052] FIG1 is a schematic flow chart of the method steps for oral CBCT ultra-low-dose imaging according to the present invention;

[0053] FIG2 is a schematic diagram of the oral CBCT image data acquisition principle of the present invention;

[0054] FIG3 is a schematic diagram of the results of the oral CBCT ultra-low-dose imaging reconstruction without image enhancement according to the present invention;

[0055] FIG4 is a schematic diagram of the principle of the image enhancement network model for oral CBCT ultra-low-dose imaging according to the present invention;

[0056] FIG5 is a schematic diagram of the image enhancement results of oral CBCT ultra-low-dose imaging according to the present invention;

[0057] FIG6 is a schematic diagram of an oral CBCT ultra-low-dose imaging system according to the present invention;

[0058] FIG7 is a schematic diagram of the connection relationship of the oral CBCT ultra-low dose imaging device of the present invention;

[0059] FIG8 is a schematic diagram showing the structural principle of the image enhancement module of the present invention.

[0060] Explanation of the accompanying symbols: 1. Oral CBCT imaging component; 2. Image reconstruction component; 3. Image display component; 4. Image enhancement component; 201. Projection acquisition module; 202. Three-dimensional reconstruction module; 203. Image enhancement module; 204. Image display and output module; 401. Memory; 402. Processor; 403. Network interface. DETAILED DESCRIPTION

[0061] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0062] As shown in FIG1 , a method for oral CBCT ultra-low-dose imaging includes the following steps:

[0063] Image data collection: collecting oral CBCT two-dimensional projection data of patients in high-dose mode;

[0064] Image data splitting processing: Splitting the collected CBCT two-dimensional projection data of the patient in high-dose mode into paired ultra-sparse sampling data and full sampling data. The ultra-sparse sampling collects 20-60 projections, and the full sampling collects 360-720 projections.

[0065] 3D reconstruction of image data: performing 3D reconstruction of the paired ultra-sparse sampling projection data and the fully sampled projection data to obtain paired ultra-sparse sampling oral CBCT reconstruction data and a fully sampled oral CBCT reconstruction data set;

[0066] Model training, using the paired ultra-sparse sampling oral CBCT reconstruction data and the fully sampled oral CBCT reconstruction data set as training samples to train a residual dense generative adversarial deep learning neural network model to obtain an oral CBCT ultra-low dose imaging image enhancement network model;

[0067] Image data input, performing three-dimensional image data reconstruction processing on the patient's oral CBCT two-dimensional projection data in ultra-low-dose mode to obtain ultra-low-dose oral CBCT reconstructed data, and inputting the data into the oral CBCT ultra-low-dose imaging image enhancement network model;

[0068] Image enhancement: the oral CBCT ultra-low-dose imaging image enhancement network model performs image enhancement processing on the ultra-low-dose oral CBCT reconstruction data to obtain a high-quality oral CBCT image of the patient;

[0069] Image data output: display and output the high-quality oral CBCT images of the patient.

[0070] Alternatively, if original 2D projection data is difficult to obtain, orthographic projection can be performed using fully reconstructed oral CBCT data to obtain multi-angle 2D projection data. Projection parameters are consistent with those of current mainstream oral CBCT equipment. In this embodiment, the source-to-detector distance is 630 mm, the X-ray source-to-object center distance is 400 mm, the flat-panel detector maximum pixel size is set to 1920 × 1536, and the detector pixel height and width are 0.15 mm.

[0071] As shown in Figure 2, each black dot on the circumference represents the location where image projection data is collected. The X-ray light source is located at the black dot on the circumference to irradiate the patient and project the patient's projection image data onto a flat-panel detector on the other side of the patient. The X-ray tube and flat-panel detector rotate 360 ​​degrees around the imaging part of the human body, capturing multi-angle two-dimensional projection data. Full sampling involves collecting 360-720 projections for reconstruction. Ultra-low-dose imaging is achieved by reducing the number of projection data collected on the rotating circumference, i.e., sparse sampling. Sparse sampling can significantly reduce scan time and imaging dose. The sparse sampling in this invention involves collecting 20-60 projections, which falls within the ultra-sparse sampling category.

[0072] Specifically, the algorithms used for the 3D reconstruction of image data include an iterative reconstruction algorithm and an FDK 3D reconstruction algorithm. In this embodiment, the CBCT iterative reconstruction algorithm is preferably used for the 3D reconstruction of image data. The reconstructed image size is 512 pixels × 512 pixels, with a pixel size of 0.15 mm × 0.15 mm.

[0073] As shown in Figure 3, the three columns of images (from left to right) show 10, 20, and 60 3D reconstructed images from 2D, 2D, and 60 3D reconstructed images of oral CBCT projection data. As shown in the figure, sparse sampling, due to a severe lack of sampling information, can easily lead to numerous artifacts in the 3D reconstructed CBCT images, severely degrading image quality. The higher the degree of sparse sampling, that is, the fewer projections, the worse the reconstructed image quality, affecting the subsequent clinical application of oral CBCT images.

[0074] As shown in Figure 4, the present invention uses a residual dense generative adversarial network (RDN-GAN) deep learning network model as the backbone network for oral CBCT ultra-low-dose imaging enhancement. The residual dense generative adversarial deep learning neural network model includes a residual dense network and an adversarial network. The residual dense network includes convolutional layers, residual dense blocks, a cascade network, upsampling, and deconvolution layers, while the adversarial network includes convolutional layers, activation functions, normalization layers, and a density network. The residual dense network serves as the generator, and the residual dense blocks include multiple convolutional layers, activation functions, cascade layers, and multiple skip connections, significantly improving network performance and accuracy. The upsampling layer is used to upsample low-resolution oral sparsely sampled CBCT reconstruction data. The adversarial network serves as the discriminator, determining whether the enhanced image is a true CBCT image. By processing the input oral sparsely sampled CBCT reconstruction data through multiple residual dense blocks, the image quality and detail can be gradually improved. Simultaneously, the discriminator determines whether the CBCT image generated by the generator approximates a true CBCT image. By repeatedly training the generator and discriminator and continuously optimizing the model, high-quality image enhancement of CBCT sparse sampling imaging is ultimately achieved.

[0075] In a specific embodiment of the present invention, the loss function of the residual dense generative adversarial deep learning neural network model is:

[0076] Among them, N is the total number of training input samples, θ G Represents the weights and biases of the residual dense generative adversarial deep learning neural network, n is the index of each training sample, l SR is the perceptual loss function, which is composed of multiple weighted loss functions. represents the output of the residual dense network, Input sample of oral sparse sampling CBCT reconstruction data for the nth training set, For paired full-sampled CBCT reconstruction data samples, is the optimal solution of the trained network model.

[0077] The network is trained to The optimal solution of , which minimizes the perceptual loss function between sparsely sampled samples and fully sampled samples.

[0078] As shown in Figure 4, in a specific embodiment of the present application, the residual dense generative adversarial deep learning neural network structure includes a residual dense network and an adversarial network; wherein, the residual dense network is composed of a convolutional layer, a residual dense block, a cascade network, an upsampling layer, and a deconvolution layer. The convolution kernel size of each layer is k×k, and each layer has a total of c channels. The adversarial network includes a convolutional layer, an activation function, a normalization layer, and a density network. The adversarial network, or discriminator, passes the input image through a convolutional layer with different parameters, a normalization layer, and an activation function to finally determine whether it is true or false.

[0079] Specifically, the residual dense network in the residual dense generative adversarial deep learning neural network model includes a convolutional layer, a residual dense block, a cascade network, an upsampling layer, and a deconvolution layer. The residual dense block is composed of a convolutional layer, an activation function, and a cascade layer. The adversarial network includes a convolutional layer, an activation function, a normalization layer, and a density network.

[0080] Among them, the input and output of each density network are: F d,c =σ(W d,c [F d-1 ,F d,1 ,…,F d,c-1 ])

[0081] Among them, F d,c is the output of the dth layer density network and the cth layer convolution network, F d-1 and F d are the input and output of the d-th layer density network, σ is the RELU activation function, W d,c is the weight of each density network convolution layer, [F d-1 ,F d,1 ,…,F d,c-1 ] is the cascade feature map generated by the d-1th layer density network, F d,1 is the output of the first convolutional network in the d-th density network, F d,c-1 is the output of the c-1th convolutional network in the dth density network.

[0082] The RELU activation function is: f(x)=max(0,x)

[0083] Among them, the RELU activation function is a linear function. Compared with the general Sigmoid or Tanh activation function, the RELU activation function does not have the problem of derivative vanishing or derivative exploding during training, which can make the entire training process more stable. In addition, the calculation of the RELU activation function is simpler and does not require floating-point operations, which greatly reduces the processing time.

[0084] The cascade network links the features of all density networks and adaptively controls the output information through a 1×1 convolutional network. Finally, the output of the entire local adversarial density network is obtained through a residual learning network.

[0085] Specifically, in order to further enhance the network capabilities, features are constructed before the RELU activation function and the perceptual loss function l is introduced. SR , and its calculation formula is:

[0086] in is the content loss function, To be the adversarial loss function,

[0087] here Represents the reconstructed image, is the probability of a real picture, represents the output of the residual dense network, I LR represents the input image of the adversarial network, θ G Represents the weights and biases of the residual dense network, N is the total number of training inputs, n is the index of each training sample, and is calculated using the MSE model and PSNR model Value: PSNR=20*log 10 (MAX I )-10*log 10 (MSE)

[0088] Among them, MAX I is the maximum value of the pixel. Since the data will be normalized during the training process, the maximum value is 1. m, n represent the horizontal and vertical pixel numbers of the input image resolution, respectively. i, j are the horizontal and vertical index numbers corresponding to each pixel. I, K represent the image after image enhancement and the fully sampled image, respectively.

[0089] Furthermore, in a specific embodiment, the model training includes:

[0090] Performing linear interpolation and affine transformation on the paired ultra-sparsely sampled oral CBCT reconstruction data and the fully sampled oral CBCT reconstruction data to normalize the images into 512 pixel × 512 pixel images, performing data expansion and normalization, and then dividing the normalized image set into a training set, a validation set, and a test set;

[0091] Specifically, the training set is used to train the RDN-GAN deep learning network model, the validation set is used to verify the effectiveness of the network model, and the test set is used to further test the effectiveness of the network model. During the training process of the RDN-GAN network model, the Peak Signal to Noise Ratio (PSNR) is used as the loss function, and the optimization algorithm adopts the Adam algorithm. During the training process, the training set data is randomly divided into small batches for training, and the number of training rounds is 5000. On the validation set and test set, the Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index (SSIM) after image enhancement are calculated to evaluate the effectiveness of the network model.

[0092] The verification model is verified and tested using the verification set and the test set. If the verification result does not meet the prediction probability threshold, the generative network and the adversarial network are retrained. If the verification result meets the prediction probability threshold, further testing is performed using the test set. If the test result meets the prediction probability threshold, the training of the oral CBCT ultra-low dose imaging image enhancement network model is completed. If the test result does not meet the prediction probability threshold, the generative network and the adversarial network are retrained.

[0093] Since all deep learning algorithm models require training, the quality and quantity of training data may affect the final model training effect. In order to achieve better results, as a specific embodiment, the training data details used by the oral CBCT ultra-low-dose imaging image enhancement model of the present invention are as follows: it contains high-dose CBCT image data of more than 1,500 patients, of which more than 200 are high-resolution imaging modes, and the imaging equipment is Newtom, Sirona and Soredex. Each patient contains approximately 200 to 500 tomographic images, with a pixel size of 0.1-0.3mm, a projection rotation angle of 360 degrees, ultra-sparse sampling for 20, 30, and 60 projection reconstructions, and full sampling reconstruction for 360 and 600 projection reconstructions.

[0094] After training is complete, all modules can retain only the test program and the trained model. In addition, fixed-point implementation is used to avoid floating-point operations, which greatly speeds up the operation of the entire system.

[0095] It should be additionally noted that the pre-trained oral CBCT ultra-low-dose imaging image enhancement network model in the present invention can be directly applied to the oral CBCT ultra-low-dose imaging image enhancement process from various equipment sources. In some cases, such as when the enhanced image quality is insufficient or there is a lot of distortion, optimization training can be performed by continuing to add oral CBCT data, especially data from current oral CBCT equipment sources, to improve the robustness and accuracy of the algorithm model.

[0096] Figure 5 shows the image enhancement results of oral CBCT ultra-low-dose imaging given by an embodiment of the present invention. The first column shows the input low-quality CBCT reconstructed image of oral CBCT ultra-low-dose imaging, with a sampling number of 20 projections. The second column shows the image enhancement results output by this embodiment of the present invention. The third column shows the ground truth of the oral CBCT high-dose (full sampling) reconstruction imaging result, which can be used as a standard for evaluating the clarity and fidelity of the image enhancement results.

[0097] As shown in Figure 5, the method of the present invention can effectively eliminate artifacts caused by extremely sparse sampling when performing CBCT image enhancement. Image quality and detail are significantly improved, achieving very high clarity and very low distortion. The enhanced image PSNR value increased from 7.84dB to 11.78dB, and the SSIM value increased from 0.34 to 0.42. Compared with existing technologies, the method of the present invention can reduce radiation dose by 95% while maintaining high imaging quality and resolution. These results demonstrate that the method of the present invention can effectively improve the quality and detail of ultra-low-dose oral CBCT imaging, providing a new solution for low-dose, high-definition oral CBCT imaging and promising application prospects in the field of oral CBCT imaging.

[0098] It should be noted that in ultra-low-dose sparse sampling of oral CBCT, the higher the sparsity, that is, the fewer projections, the more and more severe artifacts will be produced in the reconstructed image, and the greater the difficulty of the algorithm model in achieving image enhancement. After extensive trial and error, exploratory research, and model optimization, the present invention has achieved enhancement of 20-projection reconstruction images of oral CBCT. Based on extensive experiments, the present invention found that 20-projection reconstruction is close to the limit of the present invention's oral CBCT ultra-low-dose imaging method. The ability to achieve high-precision image enhancement with this extremely sparse sampling is already world-leading. Further reductions in the sampling number will result in distortion of the enhanced image. Of course, it is not ruled out that more advanced algorithms and higher-quality datasets may enable high-definition reconstruction of oral CBCT with even lower sampling numbers.

[0099] The hardware platform on which the pre-constructed oral CBCT ultra-low-dose imaging enhancement method of the present invention relies is Nvidia's GPU. On Nvidia's 4090 GPU hardware, the deep learning algorithm of the present invention can process at least 10 frames of images per second. In summary, the present invention utilizes a combination of computer vision technology and the latest artificial intelligence deep learning technology to achieve image quality enhancement of images after sparse sampling reconstruction, which can ensure the patient's imaging clarity while significantly reducing the patient's radiation dose. Overall, the present invention has the advantages of a more accurate algorithm, stronger robustness, the ability to handle more extreme situations, and the ability to effectively eliminate noise caused by various external imaging interferences.

[0100] In summary, to address the shortcomings of existing oral CBCT imaging with high doses, the present invention provides a method for oral CBCT ultra-low-dose imaging enhancement based on deep learning. This method uses the latest, more advanced residual dense generative adversarial network (RDN-GAN) deep learning neural network model to achieve image enhancement. It also uses a CBCT iterative reconstruction algorithm to perform three-dimensional reconstruction of two-dimensional projection data, enabling high-definition imaging with a lower sampling count. As shown in Figure 5, the present invention demonstrates excellent image enhancement effects. In an embodiment of the present invention, the algorithm model can very efficiently eliminate artifacts caused by sparse sampling, significantly improve image quality, enhance image details, and maintain high image fidelity. Therefore, the present invention can significantly reduce the patient's radiation dose while maintaining high imaging clarity, accelerating imaging speed, improving patient safety during examinations, and reducing artifacts related to patient motion. This method provides a new solution for low-dose, high-definition oral CBCT imaging, and has excellent application prospects in oral disease diagnosis, surgical planning, and implant surgery navigation.

[0101] As shown in FIG6 , this embodiment further provides a system for oral CBCT ultra-low-dose imaging, comprising a projection acquisition module 201 , a three-dimensional reconstruction module 202 , an image enhancement module 203 , and an image display and output module 204 connected in sequence.

[0102] The projection acquisition module 201 is used to collect oral CBCT two-dimensional projection data of the patient in ultra-low dose mode;

[0103] The three-dimensional reconstruction module 202 is used to perform three-dimensional reconstruction on the oral CBCT two-dimensional projection data in the ultra-low dose mode;

[0104] The image enhancement module 203 is used to input the oral CBCT two-dimensional projection data in the ultra-low-dose mode into the trained oral CBCT ultra-low-dose imaging image enhancement network model to obtain an image-enhanced oral CBCT image;

[0105] The image display and output module 204 is used to display the enhanced oral CBCT image and output it in Dicom file format.

[0106] After training is complete, all modules can retain only the test program and the trained model. When acquiring and processing patient images, floating-point numbers are converted to fixed-point values ​​to avoid floating-point operations, thereby increasing the speed of the entire system.

[0107] As shown in FIG7 , this embodiment further provides a device for oral CBCT ultra-low-dose imaging, comprising: an oral CBCT imaging component 1, an image reconstruction component 2, an image display component 3, and an image enhancement component 4; the oral CBCT imaging component 1 is communicatively connected to the image reconstruction component 2, and the image reconstruction component 2 is communicatively connected to the image display component 3 and the image enhancement component 4, respectively;

[0108] The image reconstruction component 2 is used to perform three-dimensional reconstruction of the data acquired by the oral CBCT imaging component 1; the image enhancement component 4 stores an oral CBCT ultra-low-dose imaging image enhancement network model, and performs image enhancement processing on the three-dimensional data reconstructed by the image reconstruction component 2; the image display component 3 is used to display the image after image enhancement processing.

[0109] Specifically, the oral CBCT imaging assembly 1 includes an X-ray tube, a flat-panel detector, a bracket, and a motion platform. The X-ray tube is arranged on the motion platform and is located on one side of the patient. The motion platform is arranged on the bracket. The flat-panel detector is arranged on the other side of the patient and matches the X-ray tube.

[0110] As shown in Figure 8 , the image enhancement component 4 includes a memory 401, a processor 402, and a network interface 403. The memory 401, processor 402, and network interface 403 are electrically connected. The memory 401 stores an image enhancement network model for ultra-low-dose oral CBCT imaging. Figure 8 is merely a block diagram of a portion of the structure related to the present application and does not limit the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figure, combine certain components, or have a different component arrangement.

[0111] This embodiment also provides a storage device for use in the oral CBCT ultra-low-dose imaging, including: phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, compact disc read-only memory (CD-ROM), or digital versatile disc (DVD).

[0112] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0113] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A method for oral CBCT ultra-low-dose imaging, characterized in that: include: Image data collection: collect the oral CBCT two-dimensional projection data of the patient in high-dose mode; Image data splitting processing, splitting the collected CBCT two-dimensional projection data of the patient in high-dose mode into paired ultra-sparse sampling projection data and full sampling projection data; The image data is processed by three-dimensional reconstruction, and the paired ultra-sparse sampling projection data and the fully sampled projection data are respectively reconstructed in three dimensions to obtain paired ultra-sparse sampling oral CBCT reconstruction data and a fully sampled oral CBCT reconstruction data set; Model training, using the paired ultra-sparse sampling oral CBCT reconstruction data and the fully sampled oral CBCT reconstruction data set as training samples to train the residual dense generative adversarial deep learning neural network model to obtain an oral CBCT ultra-low dose imaging image enhancement network model; Image data input, performing three-dimensional image data reconstruction processing on the patient's oral CBCT two-dimensional projection data in ultra-low-dose mode to obtain ultra-low-dose oral CBCT reconstruction data, and inputting the data into the oral CBCT ultra-low-dose imaging image enhancement network model; Image enhancement: the oral CBCT ultra-low-dose imaging image enhancement network model performs image enhancement processing on the ultra-low-dose oral CBCT reconstruction data to obtain a high-quality oral CBCT image of the patient; Image data output: display and output the high-quality oral CBCT image of the patient.

2. The method for oral CBCT ultra-low dose imaging according to claim 1, characterized in that: The number of projections of the ultra-sparsely sampled projection data ranges from 20 to 60 projections, and the number of projections of the fully sampled projection data ranges from 360 to 720 projections; The algorithms used in the three-dimensional reconstruction processing of the image data include a CBCT iterative reconstruction algorithm and an FDK three-dimensional reconstruction algorithm.

3. The method for oral CBCT ultra-low dose imaging according to claim 1, characterized in that: The residual dense generation adversarial deep learning neural network model includes a residual dense network and an adversarial network connected in sequence, and the loss function of the residual dense generation adversarial deep learning neural network model is: Among them, N is the total number of training input samples, θ G Represents the weights and biases of the residual densely generated adversarial deep learning neural network, n1 is the index of each training sample, l SR is the perceptual loss function, represents the output of the residual dense network, The ultra-sparse sampling oral CBCT reconstruction data input sample of the training set, The full-sampled oral CBCT reconstruction data samples paired with the training set, It is the optimal solution of the trained network model.

4. The method for oral CBCT ultra-low dose imaging according to claim 3, characterized in that: The residual dense network includes a convolution layer, a residual dense block, a cascade network, an upsampling and a deconvolution layer, and the residual dense block is composed of a convolution layer, an activation function and a cascade layer; The adversarial network consists of a convolutional layer, an activation function, a normalization layer, and a density network. The input and output of each density network are: F d,c =σ(W d,c [F d-1 ,F d,1 ,…,F d,c-1 ]) Among them, F d,c is the output of the cth layer convolutional network of the dth layer density network, F d-1 and F d are the input and output of the d-th layer density network, σ is the RELU activation function, W d,c is the weight of each convolutional layer of the density network, [F d-1 ,F d,1 ,…,F d,c-1 ] is the cascade feature map generated by the d-1th layer density network; F d,1 is the output of the first convolutional network in the dth density network, F d,c-1 is the output of the c-1th convolutional network in the dth density network.

5. The method for oral CBCT ultra-low dose imaging according to claim 4, characterized in that: Construct features before the RELU activation function and introduce the perceptual loss function l SR , the specific calculation formula is: in is the content loss function, To combat the loss function, in, Represents the reconstructed image, represents the probability of a real image, represents the output of the residual dense network, I LR represents the input image of the adversarial network, θ G Represents the weights and biases of the residual dense generative adversarial deep learning neural network, N is the total number of training input samples, n1 is the index of each training sample, and MSE and PSNR are used to calculate Values: PSNR=20*log 10 (MAX I )-10*log 10 (MSE) Among them, MAX I is the maximum value of the pixel. Since the data will be normalized during the training process, the maximum value is 1. m, n2 represent the horizontal and vertical pixel numbers of the input image resolution, respectively. i, j are the horizontal and vertical index numbers corresponding to each pixel. I, K represent the enhanced image and the fully sampled image, respectively.

6. A system for the method for oral CBCT ultra-low-dose imaging according to any one of claims 1 to 5, characterized in that: include: A projection acquisition module (201), a three-dimensional reconstruction module (202), an image enhancement module (203) and an image display and output module (204) are sequentially connected; the projection acquisition module (201) is used to collect oral CBCT two-dimensional projection data of a patient in an ultra-low dose mode; the three-dimensional reconstruction module (202) is used to perform three-dimensional reconstruction on the oral CBCT two-dimensional projection data in the ultra-low dose mode; the image enhancement module (203) is used to input the oral CBCT two-dimensional projection data in the ultra-low dose mode into an oral CBCT ultra-low dose imaging image enhancement network model to obtain an image-enhanced oral CBCT image; and the image display and output module (204) is used to display the image-enhanced oral CBCT image and output it in a Dicom file format.

7. A device applied to the method for oral CBCT ultra-low-dose imaging according to any one of claims 1 to 5, characterized in that: include: An oral CBCT imaging component (1), an image reconstruction component (2), an image display component (3) and an image enhancement component (4); the oral CBCT imaging component (1) is communicatively connected to the image reconstruction component (2), and the image reconstruction component (2) is communicatively connected to the image display component (3) and the image enhancement component (4) respectively; The image reconstruction component (2) is used to perform three-dimensional reconstruction on the data acquired by the oral CBCT imaging component (1); the image enhancement component (4) stores the oral CBCT An ultra-low-dose imaging image enhancement network model is provided, and image enhancement processing is performed on the three-dimensional data reconstructed by the image reconstruction component (2), and the image display component (3) is used to display the image after the image enhancement processing.

8. The method and device for oral CBCT ultra-low-dose imaging according to claim 7, characterized in that: The oral CBCT imaging assembly (1) comprises an X-ray tube, a flat panel detector, a bracket and a motion platform, wherein the X-ray tube is arranged on the motion platform and is located on one side of a patient, the motion platform is arranged on the bracket, and the flat panel detector is arranged on the other side of the patient and matches the X-ray tube.

9. The method and device for oral CBCT ultra-low dose imaging according to claim 7, characterized in that: The image enhancement component (4) comprises a memory (401), a processor (402) and a network interface (403), wherein the memory (401), the processor (402) and the network interface (403) are electrically connected; and an oral CBCT ultra-low-dose imaging image enhancement network model is stored in the memory (401).

10. The method and device for oral CBCT ultra-low dose imaging according to claim 9, characterized in that: The memory (401) includes: phase change memory, static random access memory, dynamic random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory, compact disc read-only memory or digital versatile disc.

Images