Method and system for ct projection data reconstruction

Abstract

The application provides a CT projection data reconstruction method and system. The method comprises the following steps: obtaining original CT projection data of a target patient, corresponding acquisition equipment and scanning scene information; analyzing the original CT projection data and the acquisition equipment and scanning scene information to obtain compliant to-be-reconstructed data and adaptive parameters; selecting a matched target CT reconstruction model from a CT reconstruction model library based on the adaptive parameters; inputting the to-be-reconstructed data into the target CT reconstruction model, and obtaining a target reconstruction image after processing by the target CT reconstruction model; performing quality control verification and standard format packaging on the target reconstruction image to obtain compliant target diagnostic image data; and sending the target diagnostic image data to an in-hospital terminal. The application can realize cloud-based and intelligent CT projection data reconstruction, thereby improving imaging quality and clinical applicability.

Description

Technical Field

[0001] This application relates to the field of medical image reconstruction technology, and in particular to a method and system for reconstructing CT projection data. Background Technology

[0002] Computed tomography (CT) is a widely used imaging technique in clinical diagnosis and treatment.

[0003] However, traditional CT imaging currently relies on local hardware and closed reconstruction algorithms. In low-dose, sparse-view scanning scenarios, the CT images reconstructed by traditional CT reconstruction algorithms are prone to artifacts, severe noise interference, and loss of details, which affect the accuracy of doctors' clinical diagnosis. Increasing the scanning dose will increase the patient's radiation exposure risk, making it difficult to balance image quality and radiation safety. Summary of the Invention

[0004] In view of this, this application provides a CT projection data reconstruction method and system to realize cloud-based and intelligent CT projection data reconstruction, thereby improving imaging quality and clinical applicability, and reducing the equipment upgrade costs of medical institutions.

[0005] In a first aspect, this application provides a CT projection data reconstruction method, applied to a cloud server of a CT projection data reconstruction system, wherein the cloud server is communicatively connected to an in-hospital terminal, and the method includes: The process involves acquiring the original CT projection data of the target patient, along with the corresponding acquisition equipment and scanning scene information. The scanning scene information characterizes the configuration status of the CT scan operation. The original CT projection data and the acquisition equipment and scanning scene information are parsed to obtain compliant data to be reconstructed and matching parameters. Based on the matching parameters, a matching target CT reconstruction model is selected from the CT reconstruction model library. The data to be reconstructed is input into the target CT reconstruction model, and after processing by the target CT reconstruction model, a target reconstructed image is obtained. The target reconstructed image undergoes quality control verification and standard format encapsulation to obtain compliant target diagnostic image data. The target diagnostic image data is sent to the in-hospital terminal so that the in-hospital terminal can display the target diagnostic image data of the target patient.

[0006] Optionally, the step of inputting the data to be reconstructed into the target CT reconstruction model, and processing it through the target CT reconstruction model to obtain the target reconstructed image, includes: The data to be reconstructed is input into the target CT reconstruction model, and the target CT reconstruction model is used to extract multi-scale projection domain features from the data to be reconstructed, and the artifact features are separated. Based on the separated artifact features, metal artifact correction and beam hardening correction in the projection domain are completed, and the artifact-corrected projection domain feature data are obtained. Based on the projection domain feature data, the mapping from the projection domain to the image domain and image reconstruction are completed to obtain the initial CT reconstruction image; Adaptive noise reduction and super-resolution enhancement are performed on the initial CT reconstruction image to obtain the target reconstruction image.

[0007] Optionally, based on the separated artifact features, the metal artifact correction and beam hardening correction in the projection domain are completed to obtain artifact-corrected projection domain feature data, including: Based on the artifact features obtained by separation, the metal projection trajectory is identified, the projection value deviation of the metal projection area is calculated, and the compensation weight matrix obtained by model learning is combined to complete the compensation of the projection value deviation of the metal artifact, and the projection data after metal artifact correction is obtained. Polynomial fitting correction was used to perform beam hardening correction on the projection data after metal artifact correction, eliminating cup-shaped and stripe-shaped artifacts caused by beam hardening, and obtaining the artifact-corrected projection domain feature data.

[0008] Optionally, the step of identifying the metal projection trajectory based on the separated artifact features, calculating the projection value deviation of the metal projection area, and compensating for the projection value deviation of the metal artifact by combining the compensation weight matrix learned from the model, to obtain the metal artifact-corrected projection data, includes: Based on the artifact features obtained from the separation, the metal projection region in the projection sine curve is located, the metal projection trajectory is fitted, and the projection value deviation corresponding to the metal projection region is calculated based on the forward projection of the metal projection region. The compensation weight matrix that matches the metal projection trajectory is obtained by learning the target CT reconstruction model. The projection value deviation is weighted and compensated based on the compensation weight matrix, and the abnormal projection value of the original metal projection area is replaced to obtain the projection data after metal artifact correction.

[0009] Optionally, the compensation weight matrix learned through the target CT reconstruction model to match the metal projection trajectory includes: Based on the adaptation parameters, the pre-trained metal artifact correction weight branch in the target CT reconstruction model is invoked. By combining the characteristics of metal artifacts in the current scan projection data, the weight branch for metal artifact correction is lightly adjusted to generate a compensation weight matrix that matches the current CT projection data and CT equipment model.

[0010] Optionally, selecting a matching target CT reconstruction model from the CT reconstruction model library based on the adaptation parameters includes: Based on the device model, scanning scenario, and scanning area information in the adaptation parameters, a model matching retrieval index is constructed. Based on the model matching retrieval index, the matching target CT reconstruction model is retrieved and filtered in the preset cloud model library, and the target CT reconstruction model is verified. The verified target CT reconstruction model is loaded onto the cloud inference computing node to complete the initial configuration of the model inference environment.

[0011] Optionally, the step of parsing the original CT projection data and the acquisition device and scanning scene information to obtain compliant data to be reconstructed and adaptation parameters includes: The original CT projection data is normalized and outlier is removed to obtain compliant data to be reconstructed. The acquisition equipment and scanning scene information are analyzed to extract at least one of the equipment model, scanning location, scanning parameters and energy spectrum characteristic parameters to generate adaptation parameters. The compliant data to be reconstructed is bound and archived with the adaptation parameters to obtain compliant data to be reconstructed and adaptation parameters that can be input into the target CT reconstruction model.

[0012] Optionally, acquiring the target patient's original CT projection data, the corresponding acquisition device, and the scanning scene information includes: Receive encrypted projection data packets transmitted from the hospital terminal, complete the signature verification and data decryption of the encrypted data packets, and restore the original CT projection data to be verified, the corresponding acquisition device to be verified, and the scanning scene information; The restored original CT projection data to be verified, the corresponding acquisition device and scanning scene information to be verified are verified, and after confirming that the data is valid, the original CT projection data of the target patient, the corresponding acquisition device and scanning scene information are obtained.

[0013] Optionally, the step of performing quality control verification and standard format encapsulation on the reconstructed target image to obtain compliant target diagnostic image data includes: The reconstructed target image is subjected to quality control verification for image sharpness, anatomical structure integrity, artifact residue, and numerical accuracy. The verified target reconstructed image is packaged according to the medical imaging standard format and embedded with the corresponding device parameters, scanning scene and reconstruction process metadata to generate compliant target diagnostic image data that can be used for clinical diagnosis.

[0014] Secondly, this application provides a CT projection data reconstruction system, which includes a cloud server and an in-hospital terminal connected by communication, wherein the cloud server includes: The data acquisition module is used to acquire the original CT projection data of the target patient, the corresponding acquisition device and scanning scene information, wherein the scanning scene information is used to characterize the configuration status of the CT scan operation; The reconstruction processing module is used to parse the original CT projection data and the acquisition device and scanning scene information to obtain compliant data to be reconstructed and adaptation parameters; select a matching target CT reconstruction model from the CT reconstruction model library based on the adaptation parameters; input the data to be reconstructed into the target CT reconstruction model, and after processing by the target CT reconstruction model, obtain the target reconstructed image; perform quality control verification and standard format encapsulation on the target reconstructed image to obtain compliant target diagnostic image data; The data output module is used to send the target diagnostic image data to the in-hospital terminal so that the in-hospital terminal can display the target diagnostic image data of the target patient.

[0015] The CT projection data reconstruction method provided in this application effectively overcomes the physical limitations of local computing power and hardware of CT equipment by deploying the reconstruction calculations and models of CT projection data on a cloud server. Furthermore, based on a reconstruction model that matches and adapts the acquisition device to the scanning scene, end-to-end processing from projection data to diagnostic images is completed through a cloud-based AI model. This effectively improves image reconstruction quality in low-dose, sparse-view scanning scenarios, thereby effectively suppressing imaging artifacts and reducing patient radiation dose while meeting the needs of high-precision clinical diagnosis. Moreover, it allows for cloud-based upgrades of the imaging capabilities of older CT equipment without replacing the local hardware, and new equipment can continuously iterate and optimize its imaging performance through cloud algorithms, thus effectively extending the overall lifespan of CT equipment and reducing investment and upgrade costs. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the steps of the CT projection data reconstruction method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the CT scanning device provided in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the working principle of the CT projection data reconstruction method provided in the embodiments of this application; Figure 4This is a schematic diagram illustrating the steps of processing a target CT reconstruction model to obtain a target reconstructed image, as provided in the embodiments of this application. Figure 5 This is a schematic diagram illustrating the working principle of obtaining a target reconstructed image by processing the target CT reconstruction model provided in the embodiments of this application; Figure 6 This is a schematic diagram comparing the original projection signal and the reconstructed projection signal provided in the embodiments of this application; Figure 7 This is a schematic diagram comparing the effects of the initial acquired CT image and the target reconstructed image provided in the embodiments of this application; Figure 8 This is a schematic diagram comparing the clinical effects of the initial lung CT image and the target reconstructed image provided in the embodiments of this application; Figure 9 This is a schematic diagram of the functional units of the cloud server provided in the embodiments of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0020] In this document, references to "embodiment" or "implementation" mean that a particular feature, structure, or characteristic described in connection with an embodiment or implementation may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] Before introducing the technical solution of this application, let's go over the technical issues in related technologies in detail.

[0022] Computed tomography (CT) is a commonly used imaging technique in modern clinical medicine, widely applied in disease screening, clinical diagnosis, and postoperative efficacy evaluation. The quality of CT images directly affects the accuracy of diagnostic results. Currently, mainstream clinical CT imaging systems typically employ a local computing mode, with image reconstruction relying on the CT device's built-in hardware and utilizing traditional, fixed, and closed reconstruction algorithms.

[0023] Furthermore, in clinical applications, to reduce radiation damage to examinees, medical institutions generally adopt low-dose scanning protocols, and to improve examination efficiency, sparse view scanning modes are also frequently used. However, traditional reconstruction algorithms are insufficient in processing projection data under low-dose, sparse view conditions. Affected by factors such as noise and missing projection information, reconstructed images are prone to various interferences such as metal artifacts, beam hardening artifacts, and stripe artifacts. At the same time, there are problems such as insufficient image clarity, blurred anatomical details, and pixel value deviations, making it difficult to meet the needs of high-precision clinical diagnosis.

[0024] While current high-performance end-to-end artificial intelligence (AI) reconstruction algorithms possess powerful artifact suppression and detail enhancement capabilities, they demand significant computing resources. Conventional CT equipment has limited local hardware configurations, typically unable to support the stable operation of high-end AI reconstruction algorithms. For older CT equipment, hardware architecture limitations prevent software upgrades from enabling iterative updates to imaging algorithms; improving image quality necessitates replacing the entire device. Even with new CT equipment equipped with high-specification local hardware, its computing power has physical limitations, hindering the full realization of advanced reconstruction algorithm performance advantages and preventing the effective release of the equipment's imaging potential.

[0025] In view of this, please refer to Figure 1 , Figure 2 and Figure 3 This application provides a method for reconstructing projection data from computed tomography (CT) imaging, applied to a cloud server 220 of a CT projection data reconstruction system 200. The cloud server 220 is communicatively connected to an in-hospital terminal 210. The method includes the following steps: S10. Acquire the original CT projection data of the target patient, the corresponding acquisition device and scanning scene information, wherein the scanning scene information is used to characterize the configuration status of the CT scan operation; parse the original CT projection data and the acquisition device and scanning scene information to obtain compliant data to be reconstructed and adaptation parameters; select a matching target CT reconstruction model from the CT reconstruction model library based on the adaptation parameters; input the data to be reconstructed into the target CT reconstruction model, and after processing by the target CT reconstruction model, obtain the target reconstructed image; perform quality control verification and standard format encapsulation on the target reconstructed image to obtain compliant target diagnostic image data.

[0026] The CT projection data reconstruction system 200 includes a cloud server 220 and an in-hospital terminal 210 connected by communication. The in-hospital terminal 210 is deployed within the hospital's clinical settings and can interact with the in-hospital CT scanning equipment 211, image archiving system, and clinical image reading terminals to upload scan data and receive and display diagnostic images. The cloud server 220 serves as the system's cloud processing platform, deploying a CT reconstruction model library and possessing data processing, model matching, and AI reconstruction functions to achieve cloud-based reconstruction of CT projection data.

[0027] The original CT projection data is generated by the hospital terminal 210 after being collected and standardized preprocessed.

[0028] The information collected includes hardware parameters such as the model of the computed tomography (CT) scanner and the specifications of the detector. The information collected includes scanning configuration parameters such as the scanning site, dose level, and view sampling rate. The information collected and the information collected are sent from the hospital terminal 210 to the cloud server 220 through the end-to-cloud secure transmission link.

[0029] The process of parsing the original CT projection data and the acquisition device and scanning scene information includes decrypting and verifying the transmitted data, removing invalid data, and obtaining compliant data to be reconstructed that is in a standard format and has reliable values. Simultaneously, key parameters are extracted from the acquisition device and scanning scene information to generate adaptation parameters for model matching, such as hardware characteristic parameters corresponding to the device model and dose and view parameters corresponding to the scanning scene.

[0030] The CT reconstruction model library stores pre-trained end-to-end AI reconstruction models adapted to different equipment models and scanning scenarios. The cloud server 220 matches a target CT reconstruction model from the library that is compatible with the current equipment and scanning scenario based on adaptation parameters. In one possible embodiment, for equipment without a dedicated model, a general-purpose basic model can be called for lightweight fine-tuning to generate a suitable target model, ensuring the effectiveness of the reconstruction process.

[0031] The target CT reconstruction model is an end-to-end artificial intelligence reconstruction model, which processes the input data to be reconstructed and sequentially performs operations such as projection domain artifact correction, image domain end-to-end reconstruction, adaptive noise reduction and super-resolution enhancement, replacing traditional filtered back projection or iterative reconstruction algorithms. This effectively suppresses artifacts and reduces noise in low-dose and sparse view scanning scenarios, and outputs high-quality target reconstruction images.

[0032] In one specific embodiment of this application, the target CT reconstruction model is built based on a deep convolutional neural network and a CT reconstruction algorithm, and adopts a multi-branch cascaded architecture. The target CT reconstruction model uses neural network architectures such as multi-scale residual networks, feature pyramid networks, or U-shaped convolutional networks (U-Net) encoder structures to achieve multi-scale projection domain feature extraction. It employs a third-order polynomial fitting correction algorithm combined with shallow convolutional correction branches to achieve beam hardening correction. It uses reconstruction methods such as filtered back-projection algorithms, joint algebraic iterative reconstruction algorithms, or differentiable filtered back-projection (FBP) convolutional networks to achieve projection domain to image domain mapping and image reconstruction. It uses deep residual networks, super-resolution convolutional neural networks, or U-Net decoder structures to achieve adaptive noise reduction and super-resolution enhancement. Furthermore, the metal artifact correction weight branch is built using architectures such as U-Net type encoder / decoder networks or residual convolutional networks.

[0033] In one possible embodiment, the quality control verification includes testing clinical diagnostic indicators such as image contrast-to-noise ratio, noise level, spatial resolution, and CT value standardization to ensure that the image meets clinical diagnostic requirements. The standard format encapsulation converts the reconstructed image into an image format conforming to medical digital imaging and communication standards, while simultaneously writing patient information, scan parameters, and reconstruction model information to generate compliant target diagnostic image data, ensuring compatibility for clinical archiving and use of the image.

[0034] S20. Send the target diagnostic image data to the in-hospital terminal 210 so that the in-hospital terminal 210 displays the target diagnostic image data of the target patient.

[0035] Specifically, the cloud server 220 sends the encrypted target diagnostic image data to the hospital terminal 210 through the end-to-cloud secure transmission link. After receiving the data, the hospital terminal 210 decrypts and parses the format to restore the standard format CT image, which is then visualized on the diagnostic terminal. At the same time, it is pushed to the hospital's medical image archiving and communication system to complete the clinical archiving for doctors to review and diagnose.

[0036] Please refer to it again. Figure 3 , Figure 3 This is a schematic diagram illustrating the working principle of the CT projection data reconstruction method provided in this application embodiment. In a specific embodiment of this application, the hospital terminal 210 includes a CT scanning device 211 and a doctor's image reading workstation 212. In the actual workflow, the CT scanning device 211 first performs a CT scan on the target subject, acquiring the original CT projection data corresponding to the target patient, and transmitting it to the doctor's image reading workstation 212 in real time. After the doctor's image reading workstation 212 completes data reception and preliminary verification, it uploads the original CT projection data to the cloud server 220 via the network. The cloud server 220 processes the received original projection data to generate target diagnostic image data. The cloud server 220 sends the processed diagnostic image data back to the hospital terminal 210, and the doctor can retrieve and view the reconstructed CT images through the doctor's image reading workstation 212 to conduct disease analysis and clinical diagnosis.

[0037] As can be seen, compared to the current situation where CT projection data reconstruction relies on the local hardware of the device for computation, resulting in poor image quality under low-dose and sparse-view scanning, and the inability of older equipment to be upgraded, the CT projection data reconstruction method provided in this application first acquires the original CT projection data of the target patient, the corresponding acquisition device and scanning scene information, and parses the original CT projection data and the acquisition device and scanning scene information to obtain compliant data to be reconstructed and adaptation parameters; based on the adaptation parameters, a matching target CT reconstruction model is selected from the CT reconstruction model library, and the data to be reconstructed is input into the target CT reconstruction model. After processing by the target CT reconstruction model, the target reconstructed image is obtained; the target reconstructed image is subjected to quality control verification and standard format encapsulation to obtain compliant target diagnostic image data, and the target diagnostic image data is sent to the hospital terminal 210 so that the hospital terminal 210 can display the target diagnostic image data of the target patient. By deploying the reconstruction calculations and models of CT projection data on a cloud server 220, the physical limitations of the local computing power and hardware of the CT equipment are effectively overcome. Furthermore, based on a reconstruction model that matches and adapts to the acquisition device and the scanning scene, end-to-end processing from projection data to diagnostic images is completed through a cloud-based AI model. This effectively improves the image reconstruction quality in low-dose, sparse-view scanning scenarios, thereby effectively suppressing imaging artifacts and reducing patient radiation dose while meeting the needs of high-precision clinical diagnosis. Moreover, the CT projection data reconstruction method provided in this application can upgrade the imaging capabilities of older equipment via the cloud without replacing the local hardware of the CT equipment. New equipment can also continuously iterate and optimize its imaging performance through cloud algorithms, thereby effectively extending the overall lifespan of the CT equipment and reducing the investment and upgrade costs.

[0038] Please see Figure 4 and Figure 5The process of inputting the data to be reconstructed into the target CT reconstruction model, and then processing it through the target CT reconstruction model to obtain the target reconstructed image, includes the following steps: S11. Input the data to be reconstructed into the target CT reconstruction model, and extract multi-scale projection domain features from the data to be reconstructed through the target CT reconstruction model to separate artifact features.

[0039] The multi-scale projection domain feature extraction involves the target CT reconstruction model using convolution operations at different scales to capture the shallow texture information and deep structural information of the data to be reconstructed, and to remove interference information from the data.

[0040] The artifact features are abnormal feature information in the data to be reconstructed caused by metal implants and uneven X-ray energy spectrum, specifically including metal artifact features and beam hardening artifact features.

[0041] As can be seen, this implementation method achieves the separation of artifact features through multi-scale feature extraction, which can avoid the confusion between artifact information and normal tissue features, prevent artifacts from being solidified into the image during the reconstruction process, and effectively improve the purity of image reconstruction.

[0042] Please see Figure 6 , Figure 6 This is a schematic diagram comparing the original projection signal and the reconstructed projection signal provided in an embodiment of this application. In one possible embodiment of this application, signal optimization processing is performed on the data to be reconstructed. Figure 6 Part (a) is the original projection signal obtained by low-dose CT scan in the comparative embodiment of this application. Affected by factors such as scanning noise, beam hardening disturbance and detector sampling deviation, the signal curve is full of a large number of spikes and random fluctuations, and the waveform contour is messy and discrete. If CT image reconstruction is carried out directly based on the distorted projection signal, the final CT image will produce obvious strip artifacts and particle noise, resulting in a decrease in the image detail recognition. Figure 6 Part (b) is the reconstructed projection signal after signal optimization processing. Random noise in the original signal is effectively filtered out, distortion deviation is compensated and corrected, and the signal curve is continuous and smooth. Based on the reconstructed projection signal after correction processing, the projection domain to image domain is mapped and reconstructed, which can effectively suppress low-dose CT imaging artifacts, thereby improving the clarity and structural restoration accuracy of the final CT diagnostic image.

[0043] S12. Based on the separated artifact features, complete the metal artifact correction and beam hardening correction in the projection domain to obtain the artifact-corrected projection domain feature data.

[0044] The metal artifact correction is for strip-shaped artifacts caused by metal implants. It locates the metal projection area by artifact features and completes projection value compensation and repair.

[0045] The beam hardening correction addresses cup-shaped and stripe-shaped artifacts caused by the attenuation of the X-ray energy spectrum, correcting the projection data through polynomial fitting. The artifact-corrected projection domain feature data is projection data that eliminates artifact interference and retains only the attenuation information of human tissue.

[0046] Understandably, metal artifacts and beam hardening artifacts are the main imaging interferences in low-dose, sparse-view scanning scenarios. By correcting these two types of artifacts, abnormal interferences in the projection data can be eliminated, restoring the true tissue projection information.

[0047] S13. Based on the projection domain feature data, complete the mapping from the projection domain to the image domain and image reconstruction to obtain the initial CT reconstruction image.

[0048] Understandably, the projection domain is the integral data of rays penetrating the human body, which cannot be directly used for clinical diagnosis. It needs to be mapped and converted into image domain data in order to form a visualized tomographic image.

[0049] The mapping from the projection domain to the image domain involves coordinate transformation and numerical mapping to convert two-dimensional projection data into a two-dimensional tomographic image. A differentiable mapping algorithm is used to convert the projection data into image pixels. The initial CT reconstruction image is the base image after the domain mapping is completed, restoring the basic structural information of human tissue.

[0050] S14. Adaptive noise reduction and super-resolution enhancement are performed on the initial CT reconstruction image to obtain the target reconstruction image.

[0051] In one possible embodiment, the adaptive noise reduction dynamically adjusts the noise reduction intensity based on the noise distribution characteristics of local areas of the image, so as to preserve tissue edges and lesion details while suppressing quantum noise and Poisson noise.

[0052] In one possible embodiment, the super-resolution enhancement is achieved by improving the spatial resolution of the image through subpixel convolution operations, thereby restoring minute lesions and fine tissue structures. The target reconstructed image is a low-noise, high-resolution clinical diagnostic-grade image.

[0053] Please see Figure 7 , Figure 7 This is a schematic diagram comparing the effects of the initially acquired CT image and the target reconstructed image provided in an embodiment of this application. Figure 7The middle (c) region is the initial acquired image obtained by direct reconstruction of CT scan in low-dose mode in the comparative embodiment of this application. Affected by factors such as low scan dose, beam hardening effect, and metal interference, the image is filled with a large number of oblique lines to represent random noise and strip artifacts. The internal details of the tissue are obscured by noise, and the boundaries of different anatomical regions are blurred, which is not conducive to doctors distinguishing fine tissue structures and is difficult to meet the requirements of accurate clinical diagnosis. Figure 7 The middle (d) region is the target reconstructed image obtained after the target CT reconstruction model of this application has completed projection domain artifact correction, image reconstruction, adaptive noise reduction and super-resolution enhancement. Various noises and strip artifacts in the original image are effectively suppressed and removed, and only continuous and accurate structural contour lines are retained. The boundaries of each chamber are sharp and distinct, and the tissue structure recognition is effectively improved.

[0054] As can be seen, the CT projection data reconstruction method provided in this application first separates artifact features through multi-scale feature extraction in the projection domain, and completes dual correction of metal artifacts and beam hardening. Then, it completes the mapping and basic reconstruction from the projection domain to the image domain. Finally, it suppresses scanning noise while preserving anatomical structure and lesion details through adaptive noise reduction, and improves the spatial resolution of the image through super-resolution enhancement. This method can effectively improve the imaging quality in low-dose, sparse view scanning scenarios, avoid artifact residue and detail blurring, and enable the clarity, anatomical structure integrity and numerical accuracy of the target reconstructed image to meet the clinical diagnostic standard.

[0055] Optionally, the step of performing metal artifact correction and beam hardening correction in the projection domain based on the separated artifact features to obtain artifact-corrected projection domain feature data includes the following steps: S121. Based on the artifact features obtained from the separation, identify the metal projection trajectory, calculate the projection value deviation of the metal projection area, and combine the compensation weight matrix obtained by the model learning to complete the compensation of the projection value deviation of the metal artifact, and obtain the projection data after metal artifact correction.

[0056] The metal projection trajectory is a continuous abnormal projection path formed by the metal implant in the projection data, which is located by the metal artifact feature in the artifact features. The projection value deviation is the difference between the actual collected projection value and the true tissue projection value in the metal projection area, which is the numerical cause of metal artifacts.

[0057] The compensation weight matrix is ​​a set of weight parameters learned by the target CT reconstruction model through pre-training with clinical data. It is matched with the metal projection trajectory and used to compensate for the deviation.

[0058] It is understandable that metal implants can cause numerical deviations in the projection data, and direct reconstruction will result in obvious strip-shaped metal artifacts. This implementation method can restore the true projection values ​​of the metal area through trajectory recognition, deviation calculation and weight compensation, thereby effectively eliminating the interference of metal artifacts.

[0059] S122. The beam hardening correction is performed on the projection data after metal artifact correction using a polynomial fitting correction method to eliminate cup-shaped artifacts and strip-shaped artifacts caused by beam hardening, and the artifact-corrected projection domain feature data is obtained.

[0060] The polynomial fitting correction is based on a polynomial function of a preset order, which performs numerical fitting correction on the projection data after metal artifact correction to adapt to the X-ray energy spectrum attenuation characteristics of the CT equipment. The cup-shaped artifact and the strip-shaped artifact are imaging artifacts caused by beam hardening, which can lead to image grayscale distortion and blurred edges, affecting the accuracy of clinical diagnosis.

[0061] It is understandable that when rays pass through human tissue, spectral hardening occurs, leading to systematic deviations in the projection data. Polynomial fitting correction can specifically correct these deviations and eliminate various artifacts caused by beam hardening.

[0062] Please see Figure 8 , Figure 8 This is a schematic diagram comparing the clinical effects of an initial lung CT image and a target reconstructed image provided in an embodiment of this application. Figure 8 The middle (e) region is the initial CT projection image directly reconstructed in the low-dose scanning mode in the comparative embodiment of this application. Affected by the reduced scanning dose and beam hardening effect, the image background is full of particulate noise, and obvious horizontal strip artifacts can be seen in the lung field area. The position marked by the lead line in the figure is a typical artifact feature. Artifacts can cover up small lesions such as small bronchi and micro nodules, interfering with doctors' interpretation. At the same time, noise will reduce the overall image signal-to-noise ratio. Figure 8 The middle (f) region is the target reconstructed image obtained after projection domain artifact correction, image reconstruction, adaptive noise reduction and super-resolution enhancement in the embodiment of this application. The original strip artifacts and background noise are effectively removed, the lung parenchyma, the demarcation of the layers of bronchi and mediastinal tissues are clear and sharp, and the identification of small lesions is effectively improved.

[0063] As can be seen, the CT projection data reconstruction method provided in this application first identifies the metal projection trajectory based on the separated artifact features, calculates the projection value deviation of the metal projection region, and completes the compensation of the projection value deviation of the metal artifact by combining the compensation weight matrix learned by the model, thus obtaining the metal artifact-corrected projection data. Then, a polynomial fitting correction method is used to complete the beam hardening correction of the metal artifact-corrected projection data, eliminating the cup-shaped artifact and strip-shaped artifact caused by beam hardening, thus obtaining the artifact-corrected projection domain feature data. Since the metal projection trajectory is located and the projection value deviation is quantified based on the artifact features first, and the numerical compensation of the metal artifact is achieved through the compensation weight matrix learned by the model, the projection anomaly caused by the metal implant is effectively corrected. Then, the polynomial fitting correction is used to adapt to the attenuation characteristics of the X-ray energy spectrum, specifically eliminating the cup-shaped artifact and strip-shaped artifact caused by beam hardening. Therefore, it can achieve dual correction of metal artifacts and beam hardening artifacts, avoid artifact residue and image grayscale distortion, thereby improving the purity and accuracy of the projection data, and further ensuring the anatomical integrity and clinical diagnostic usability of the reconstructed image.

[0064] Optionally, the step of identifying the metal projection trajectory based on the separated artifact features, calculating the projection value deviation of the metal projection area, and compensating for the projection value deviation of the metal artifact by combining the compensation weight matrix learned by the model, to obtain the metal artifact-corrected projection data, includes the following steps: S1211. Based on the artifact features obtained from the separation, locate the metal projection region in the projection sine diagram, fit the metal projection trajectory, and calculate the projection value deviation corresponding to the metal projection region based on the forward projection of the metal projection region.

[0065] The projection sine curve is a two-dimensional line integral projection dataset generated after the original projection data has been standardized and preprocessed, and the metal projection region is the high-contrast abnormal pixel region corresponding to the metal implant in the projection sine curve.

[0066] The metal projection trajectory is a continuous trajectory obtained by curve fitting of the positioned metal projection area, which represents the distribution pattern of the metal implant in the projection domain.

[0067] The forward projection calculation is a forward extrapolation calculation of the metal projection area based on the scanning geometric parameters, used to quantify the difference between the actual collected value and the theoretical true value. The resulting projection value deviation is used to reflect the degree of numerical deviation of the metal artifact.

[0068] It is understandable that metal implants will exhibit a regular and continuous abnormal trajectory in the projected sine curve. This implementation method can effectively quantify the projection numerical deviation of the metal projection area through positioning, fitting and forward projection calculation.

[0069] S1212. The compensation weight matrix that matches the metal projection trajectory is obtained by learning the target CT reconstruction model. The projection value deviation is weighted and compensated based on the compensation weight matrix to replace the abnormal projection value of the original metal projection area, and the projection data after metal artifact correction is obtained.

[0070] The compensation weight matrix is ​​obtained by training the target CT reconstruction model based on clinical metal artifact samples and is adapted to metal projection trajectories of different shapes and locations. The weighted compensation involves multiplying the projection value deviation by its corresponding weight and then adding the result to the original projection value to correct abnormal values. Abnormal projection value replacement involves overwriting the original abnormal values ​​with corrected standard projection values, thereby restoring the true tissue projection information of the metal region.

[0071] Understandably, a single deviation calculation cannot achieve accurate correction. This implementation method combines the adaptive weight matrix learned by the target CT reconstruction model for weighted compensation, which can adapt to the artifact characteristics of different metal materials and different implantation sites, ensuring that the correction results conform to the actual attenuation characteristics of human tissue.

[0072] As can be seen, the CT projection data reconstruction method provided in this embodiment locates the metal projection region in the projection sine curve based on the separated artifact features, fits the metal projection trajectory, and calculates the projection value deviation corresponding to the metal projection region based on the forward projection calculation. A compensation weight matrix matching the metal projection trajectory is learned through the target CT reconstruction model. The projection value deviation is then weighted and compensated based on the compensation weight matrix to replace the abnormal projection values ​​of the original metal projection region, resulting in metal artifact-corrected projection data. Because the metal projection region in the projection sine curve is first located based on artifact features and a continuous projection trajectory is fitted, the metal projection value deviation is quantified through forward projection calculation, and then a compensation weight matrix adapted to the metal projection trajectory is generated using model learning to complete the weighted compensation of the deviation and replacement of abnormal projection values, this method can effectively correct the projection value distortion caused by metal implants, avoid artifact residue or tissue detail distortion caused by compensation deviation, thereby improving the accuracy and reliability of metal artifact correction and preserving the true projection information of normal human tissue.

[0073] Optionally, the step of learning the compensation weight matrix that matches the metal projection trajectory through the target CT reconstruction model includes the following steps: S121a. Based on the adaptation parameters, call the pre-trained metal artifact correction weight branch in the target CT reconstruction model.

[0074] The adaptation parameters include equipment and scene-related configuration information such as CT equipment model, detector specifications, scanning area and dose level.

[0075] The metal artifact correction weight branch is a sub-network branch in the target CT reconstruction model used for metal artifact compensation calculation. It is pre-trained from clinical metal artifact sample data of multiple device types and multiple scanning scenarios and has the ability to adapt and calculate metal artifact correction weights.

[0076] It is understandable that different CT scanner models have hardware differences in X-ray output characteristics and detector acquisition response, and using general weight parameters usually cannot adapt to the imaging characteristics of the current equipment. This implementation method completes the invocation of the metal artifact correction weight branch based on the adaptation parameters, which can effectively match the weight model adapted to the current equipment and scanning scene.

[0077] S121b: Combining the metal artifact characteristics of the current scan projection data, the metal artifact correction weight branch is lightly adjusted to generate a compensation weight matrix that matches the current CT projection data and CT equipment model.

[0078] The lightweight fine-tuning involves iteratively optimizing the core parameters of the metal artifact correction weight branch within a small range, without needing to train the entire model. This improves processing efficiency while ensuring correction accuracy. The compensation weight matrix generated through lightweight fine-tuning can adapt to the morphology and distribution characteristics of metal artifacts in the current projection data, while also matching the hardware imaging characteristics of the corresponding CT equipment.

[0079] In one possible embodiment, the lightweight fine-tuning involves iteratively optimizing the metal artifact feature weight coefficient, projection bias compensation coefficient, trajectory fitting correction coefficient, and energy spectrum attenuation adaptation coefficient of the metal artifact correction weight branch.

[0080] As can be seen, the CT projection data reconstruction method provided in this embodiment, based on adaptation parameters, calls the pre-trained metal artifact correction weight branch in the target computed tomography (CT) reconstruction model; combined with the metal artifact characteristics of the current scan projection data, it performs lightweight fine-tuning of the metal artifact correction weight branch, generating a compensation weight matrix that matches the current CT projection data and the CT scanner model. Since the method first calls the pre-trained weight branch matching the device and scene according to the adaptation parameters, and then performs lightweight fine-tuning based on the metal artifact characteristics of the current projection data, it can achieve rapid optimization of model parameters, generate a compensation weight matrix that matches the current scan data and device model, improve the adaptation accuracy between the weights and the metal projection trajectory, thereby making the metal artifact correction more consistent with the actual imaging characteristics and enhancing the stability and accuracy of metal artifact correction.

[0081] Optionally, selecting a matching target CT reconstruction model from the CT reconstruction model library based on the adaptation parameters includes the following steps: S15. Based on the device model, scanning scene and scanning part information in the adaptation parameters, construct a model matching retrieval index.

[0082] The device model is used to characterize the specifications, hardware architecture, detector parameters, and inherent imaging characteristics of the CT acquisition equipment. The scanning scene is used to characterize the clinical configuration of this scan, including different scanning modes such as low-dose scanning, conventional-dose scanning, sparse-view scanning, and full-view scanning. The scanning site information is used to characterize the human anatomical scanning area corresponding to this CT scan, including the head, chest, abdomen, spine, and limbs.

[0083] The model matching retrieval index is a retrieval identifier formed by integrating the above feature information, used to locate the reconstruction model that is compatible with the current device and scanning conditions.

[0084] S16. Based on the model matching retrieval index, retrieve and filter the matching target CT reconstruction model in the preset cloud model library, and verify the target CT reconstruction model.

[0085] The preset cloud model library stores pre-trained reconstruction models adapted to different specifications of CT equipment and different scanning scenarios.

[0086] Model verification includes model weight verification, model version verification, and model-device-scene compatibility verification, which are used to exclude damaged or incompatible model files.

[0087] S17. Load the verified target CT reconstruction model to the cloud inference computing node to complete the initial configuration of the model inference environment.

[0088] The cloud-based inference computing nodes are computing power execution units allocated by the cloud for artificial intelligence reconstruction tasks. Model inference environment initialization includes loading model operators, configuring runtime parameters, binding computing resources, and adapting data interfaces, enabling the target reconstruction model to enter a runnable state.

[0089] As can be seen, the CT projection data reconstruction method provided in this embodiment first constructs a retrieval index through adaptation parameters to achieve rapid model matching; then, through model compliance and adaptability verification, abnormal and mismatched models are eliminated to ensure the stability of model operation; finally, the model loading and inference environment standardization initialization are completed, which can achieve accurate and efficient matching of the reconstruction model, so that the target CT reconstruction model can effectively fit the imaging characteristics of the current equipment model, scanning scene and scanning part, and ensure the stability and reliability of the reconstruction effect.

[0090] Optionally, the step of parsing the original CT projection data and the acquisition device and scanning scene information to obtain compliant data to be reconstructed and adaptation parameters includes the following steps: S18. The original CT projection data is normalized and outliers are removed to obtain compliant data to be reconstructed. The acquisition equipment and scanning scene information are analyzed to extract at least one of the equipment model, scanning location, scanning parameters and energy spectrum characteristic parameters to generate adaptation parameters.

[0091] The normalization process involves standardizing and adjusting the numerical dimensions and data format of the original projection data to conform to the input format requirements of the target reconstruction model. The outlier removal process identifies and removes invalid values ​​from the original projection data caused by acquisition interference or equipment fluctuations, preventing abnormal data from interfering with the reconstruction results.

[0092] The scanning parameters are the operating condition configuration parameters corresponding to this CT scan, specifically including parameters such as scanning tube current, scanning tube voltage, slice thickness, scanning field of view, rotation cycle, number of projection views, scanning dose level, and sampling resolution.

[0093] The energy spectrum characteristic parameters are energy spectrum distribution related characteristic parameters of CT scan X-rays, including effective photon energy, energy spectrum half width at half maximum, radiation hardening baseline and energy attenuation coefficient distribution, etc., which are used to characterize the energy attenuation change law after X-rays penetrate human tissue.

[0094] S19. Bind and archive the compliant data to be reconstructed with the adaptation parameters to obtain compliant data to be reconstructed and adaptation parameters that can be input into the target CT reconstruction model.

[0095] The binding and archiving process involves establishing a relationship between compliant data to be reconstructed and adaptation parameters corresponding to the same patient and the same scan, and completing data storage and identification recording.

[0096] As can be seen, the CT projection data reconstruction method provided in this application first unifies the projection data format and removes abnormal values ​​to eliminate interference from invalid data. Then, it generates adaptation parameters through multi-dimensional information analysis. Finally, it achieves the correspondence between data and parameters by binding and archiving, effectively avoiding reconstruction distortion caused by data anomalies, thereby further optimizing the accuracy of cloud reconstruction.

[0097] Optionally, acquiring the target patient's original CT projection data, the corresponding acquisition device, and the scanning scene information includes the following steps: S20: Receive the encrypted projection data packet transmitted by the hospital terminal 210, complete the signature verification and data decryption of the encrypted data packet, and restore the original CT projection data to be verified, the corresponding acquisition device to be verified, and the scanning scene information.

[0098] The encrypted projection data packet is a transmission data packet generated by the in-hospital terminal 210 after encrypting the original projection data, acquisition equipment, and scanning scene information. The signature verification is to verify the legality of the digital signature of the data packet and confirm that the data source is a compliant in-hospital terminal 210.

[0099] S21. Verify the restored original CT projection data to be verified, the corresponding acquisition device and scanning scene information to be verified, and after confirming that the data is valid, obtain the original CT projection data of the target patient, the corresponding acquisition device and scanning scene information.

[0100] The data verification includes data integrity verification, format compliance verification, and numerical validity verification.

[0101] As can be seen, the CT projection data reconstruction method provided in this application first transmits data through encrypted data packet transmission, signature verification, and data decryption to ensure the security of patient privacy data and clinical imaging data transmission. Then, it filters out valid original data through multi-dimensional data verification and removes abnormal or invalid data, which can ensure the authenticity and credibility of the original projection data and supporting information entering the reconstruction process. This effectively avoids data security risks and input anomaly problems, and thus avoids reconstruction failure and imaging distortion caused by missing or invalid data.

[0102] Optionally, the step of performing quality control verification and standard format encapsulation on the reconstructed target image to obtain compliant target diagnostic image data includes the following steps: S22. Perform quality control verification on the reconstructed target image for image sharpness, anatomical structure integrity, artifact residue, and numerical accuracy.

[0103] Among them, image sharpness verification is used to detect the spatial resolution and detail restoration capability of the target reconstructed image, confirming that the image can clearly present minute anatomical structures and lesion features; anatomical structure integrity verification is used to check whether the image completely covers the morphology of the tissues and organs of the scanned area, without structural missing, tomographic distortion or edge distortion; artifact residue verification is used to determine whether interference factors such as metal artifacts and beam hardening artifacts have been eliminated; numerical accuracy verification is used to calibrate the radiation attenuation numerical standard of the image to ensure that the value conforms to the true law of human tissue attenuation.

[0104] S23. The verified target reconstructed image is packaged according to the medical imaging standard format, and the corresponding device parameters, scanning scene and reconstruction process metadata are embedded to generate compliant target diagnostic image data that can be used for clinical diagnosis.

[0105] The reconstruction process metadata refers to the process record data generated by the cloud image reconstruction process, which records the process information and parameter information of image reconstruction, such as the target CT reconstruction model model called for reconstruction, model fine-tuning parameters, artifact correction parameters, data processing timestamps, and quality control verification results.

[0106] As can be seen, the CT projection data reconstruction method provided in this application first undergoes quality control verification to eliminate substandard images, ensuring that the clarity, structural integrity, and numerical accuracy of the reconstructed images meet clinical diagnostic standards. Then, it completes the format encapsulation and embeds metadata according to medical imaging standards, enabling the images to have good device compatibility and process traceability. It can generate compliant diagnostic images that can be used for clinical image reading and archiving, adapting to the existing clinical workflow within the hospital. At the same time, it effectively avoids clinical use obstacles caused by substandard image quality and format incompatibility, improving the clinical practicality and standardization of cloud-reconstructed images.

[0107] Please see Figure 9 This application also provides an application for, for example Figure 3 The cloud server 220 of the CT projection data reconstruction system 200 shown includes: The data acquisition module 221 is used to acquire the original CT projection data of the target patient, the corresponding acquisition device and scanning scene information, wherein the scanning scene information is used to characterize the configuration status of the CT scan operation; The reconstruction processing module 222 is used to parse the original CT projection data and the acquisition device and scanning scene information to obtain compliant data to be reconstructed and adaptation parameters; select a matching target CT reconstruction model from the CT reconstruction model library based on the adaptation parameters; input the data to be reconstructed into the target CT reconstruction model, and after processing by the target CT reconstruction model, obtain the target reconstructed image; perform quality control verification and standard format encapsulation on the target reconstructed image to obtain compliant target diagnostic image data; The data output module 223 is used to send the target diagnostic image data to the in-hospital terminal 210 so that the in-hospital terminal 210 can display the target diagnostic image data of the target patient.

[0108] The CT projection data reconstruction system 200 includes, but is not limited to, public cloud deployment architecture, private cloud local data center deployment architecture, hybrid cloud collaborative deployment architecture, or edge-cloud integrated deployment architecture. The cloud server 220 is used to receive, parse, reconstruct, perform quality control packaging, and output projection data, and is used to execute the CT projection data reconstruction method provided in any embodiment of this application. The in-hospital terminal 210 is deployed on the hospital's intranet and is used for raw data acquisition, transmission, and diagnostic image display, working in conjunction with the cloud server 220 to complete the intelligent reconstruction of CT projection data.

[0109] As can be seen, the CT projection data reconstruction system 200 provided in this application deploys the reconstruction calculation and model of CT projection data on a cloud server 220, effectively overcoming the physical limitations of the local computing power and hardware of the CT equipment. Simultaneously, based on a reconstruction model matched and adapted to the acquisition device and scanning scene, it completes end-to-end processing from projection data to diagnostic images through a cloud-based AI model. Therefore, it can effectively improve the image reconstruction quality in low-dose, sparse-view scanning scenarios, thereby effectively suppressing imaging artifacts and reducing patient radiation dose while meeting the needs of high-precision clinical diagnosis. Furthermore, the CT projection data reconstruction system 200 provided in this application can achieve cloud-based upgrades of the imaging capabilities of older equipment without replacing the local hardware of the CT equipment. New equipment can also continuously iterate and optimize its imaging performance through cloud algorithms, thereby effectively extending the overall lifespan of the CT equipment and reducing the investment and upgrade costs of CT equipment.

[0110] In this application, the terms "embodiment" and "implementation" mean that a specific feature, structure, or characteristic described in connection with an embodiment can be included in at least one embodiment of this application. The appearance of these phrases in various locations throughout the specification does not necessarily refer to the same embodiment, nor are they independent or alternative embodiments mutually exclusive with other embodiments. Those skilled in the art will understand, explicitly and implicitly, that the embodiments described in this application can be combined with other embodiments. Furthermore, it should be understood that the features, structures, or characteristics described in the various embodiments of this application can be arbitrarily combined to form another embodiment that does not depart from the spirit and scope of the technical solution of this application, provided there is no contradiction between them.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to the above preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of this application should not depart from the spirit and scope of the technical solutions of this application.

Claims

1. A method for reconstructing CT projection data, characterized in that, A cloud server used in a CT projection data reconstruction system, wherein the cloud server is communicatively connected to an in-hospital terminal, the method comprising: The process involves: acquiring the original CT projection data of the target patient, the corresponding acquisition equipment, and scanning scene information, where the scanning scene information characterizes the configuration status of the CT scan operation; parsing the original CT projection data and the acquisition equipment and scanning scene information to obtain compliant data to be reconstructed and adaptation parameters; selecting a matching target CT reconstruction model from a CT reconstruction model library based on the adaptation parameters; inputting the data to be reconstructed into the target CT reconstruction model, and processing it through the target CT reconstruction model to obtain the target reconstructed image, including: inputting the data to be reconstructed into the target CT reconstruction model, and performing multi-scale projection domain analysis on the data to be reconstructed through the target CT reconstruction model. Feature extraction and separation are performed to obtain artifact features, which are abnormal feature information in the data to be reconstructed caused by metal implants and uneven X-ray energy spectrum, including metal artifact features and beam hardening artifact features. Based on the separated artifact features, metal artifact correction and beam hardening correction are performed in the projection domain to obtain artifact-corrected projection domain feature data. Based on the projection domain feature data, the mapping from the projection domain to the image domain and image reconstruction are performed to obtain the initial CT reconstruction image. Adaptive noise reduction and super-resolution enhancement are performed on the initial CT reconstruction image to obtain the target reconstructed image. The target reconstructed image is subjected to quality control verification and standard format packaging to obtain compliant target diagnostic image data. The target diagnostic image data is sent to the in-hospital terminal so that the in-hospital terminal can display the target diagnostic image data of the target patient.

2. The method as described in claim 1, characterized in that, Based on the separated artifact features, metal artifact correction and beam hardening correction are performed in the projection domain to obtain artifact-corrected projection domain feature data, including: Based on the artifact features obtained by separation, the metal projection trajectory is identified, the projection value deviation of the metal projection area is calculated, and the compensation weight matrix obtained by model learning is combined to complete the compensation of the projection value deviation of the metal artifact, and the projection data after metal artifact correction is obtained. Polynomial fitting correction was used to perform beam hardening correction on the projection data after metal artifact correction, eliminating cup-shaped and stripe-shaped artifacts caused by beam hardening, and obtaining the artifact-corrected projection domain feature data.

3. The method as described in claim 2, characterized in that, The process involves identifying the metal projection trajectory based on the separated artifact features, calculating the projection value deviation of the metal projection region, and compensating for the projection value deviation of the metal artifact using a compensation weight matrix learned from the model. This results in corrected projection data, including: Based on the artifact features obtained from the separation, the metal projection region in the projection sine curve is located, the metal projection trajectory is fitted, and the projection value deviation corresponding to the metal projection region is calculated based on the forward projection of the metal projection region. The compensation weight matrix that matches the metal projection trajectory is obtained by learning the target CT reconstruction model. The projection value deviation is weighted and compensated based on the compensation weight matrix, and the abnormal projection value of the original metal projection area is replaced to obtain the projection data after metal artifact correction.

4. The method as described in claim 3, characterized in that, The compensation weight matrix, learned through the target CT reconstruction model to match the metal projection trajectory, includes: Based on the adaptation parameters, the pre-trained metal artifact correction weight branch in the target CT reconstruction model is invoked. Based on the metal artifact characteristics of the current scan projection data, the metal artifact correction weight branch is lightly adjusted to generate a compensation weight matrix that matches the current CT projection data and CT equipment model.

5. The method according to any one of claims 1-4, characterized in that, The step of selecting a matching target CT reconstruction model from the CT reconstruction model library based on the adaptation parameters includes: Based on the device model, scanning scenario, and scanning area information in the adaptation parameters, a model matching retrieval index is constructed. Based on the model matching retrieval index, the matching target CT reconstruction model is retrieved and filtered in the preset cloud model library, and the target CT reconstruction model is verified. The verified target CT reconstruction model is loaded onto the cloud inference computing node to complete the initial configuration of the model inference environment.

6. The method according to any one of claims 1-4, characterized in that, The process of parsing the original CT projection data and the acquisition device and scanning scene information to obtain compliant data to be reconstructed and adaptation parameters includes: The original CT projection data is normalized and outlier is removed to obtain compliant data to be reconstructed. The acquisition equipment and scanning scene information are analyzed to extract at least one of the equipment model, scanning location, scanning parameters and energy spectrum characteristic parameters to generate adaptation parameters. The compliant data to be reconstructed is bound and archived with the adaptation parameters to obtain compliant data to be reconstructed and adaptation parameters that can be input into the target CT reconstruction model.

7. The method according to any one of claims 1-4, characterized in that, The acquisition of the target patient's original CT projection data, corresponding acquisition equipment, and scanning scene information includes: Receive encrypted projection data packets transmitted from the hospital terminal, complete the signature verification and data decryption of the encrypted data packets, and restore the original CT projection data to be verified, the corresponding acquisition device to be verified, and the scanning scene information; The restored original CT projection data to be verified, the corresponding acquisition device and scanning scene information to be verified are verified, and after confirming that the data is valid, the original CT projection data of the target patient, the corresponding acquisition device and scanning scene information are obtained.

8. The method according to any one of claims 1-4, characterized in that, The process of performing quality control verification and standard format encapsulation on the reconstructed target image to obtain compliant target diagnostic image data includes: The reconstructed target image is subjected to quality control verification for image sharpness, anatomical structure integrity, artifact residue, and numerical accuracy. The verified target reconstructed image is packaged according to the medical imaging standard format and embedded with the corresponding device parameters, scanning scene and reconstruction process metadata to generate compliant target diagnostic image data that can be used for clinical diagnosis.

9. A CT projection data reconstruction system, characterized in that, The CT projection data reconstruction system includes a cloud server and an in-hospital terminal connected by communication. The cloud server includes: The data acquisition module is used to acquire the original CT projection data of the target patient, the corresponding acquisition device and scanning scene information, wherein the scanning scene information is used to characterize the configuration status of the CT scan operation; The reconstruction processing module is used to parse the original CT projection data and the acquisition device and scanning scene information to obtain compliant data to be reconstructed and adaptation parameters; select a matching target CT reconstruction model from the CT reconstruction model library based on the adaptation parameters; input the data to be reconstructed into the target CT reconstruction model, and after processing by the target CT reconstruction model, obtain the target reconstructed image; perform quality control verification and standard format encapsulation on the target reconstructed image to obtain compliant target diagnostic image data; The data output module is used to send the target diagnostic image data to the in-hospital terminal so that the in-hospital terminal can display the target diagnostic image data of the target patient.

Images