X-ray detector array imaging system

Abstract

The invention discloses an X-ray detector array imaging system, which relates to the technical field of array imaging, and is based on three-dimensional tensor input constructed by continuous frames and combined with a generation mechanism of an acceleration channel, so that the system can sensitively capture sudden displacement changes in an operation, such as non-uniform motion caused by sudden arrest of an electric saw or respiratory muscle fibrillation. The curvature change of a displacement curve is accurately quantified through Gaussian weighted second-order difference, weak jitter interference is filtered through an adaptive suppression coefficient, and high-robustness time-varying characteristics are provided for motion compensation; a spatial feature extraction branch of the double-branch processing unit retains anatomical details of bones and blood vessels, and a time motion modeling branch analyzes a long-term dependency relationship of a displacement field sequence through an LSTM network; the two are spliced through a fusion layer and then input into an adversarial training module, and under a generator-discriminator game mechanism, not only is the overall motion artifact corrected, but also the microstructure loss caused by excessive smoothness is avoided.

Description

Technical Field

[0001] This invention relates to the field of array imaging technology, and in particular to an X-ray detector array imaging system. Background Technology

[0002] In minimally invasive orthopedic and thoracic surgeries, C-arm X-ray imaging systems need to continuously capture images from multiple angles around the patient to guide the positioning of surgical instruments in real time. During the operation, the patient's unconscious muscle tremors and respiratory fluctuations under anesthesia, as well as the sudden displacement of instruments operated by the doctor, cause irregular movements of the imaging target within millisecond intervals. This complex movement includes low-frequency tremors and high-frequency sudden stops, which exceeds the compensation range of conventional uniform motion models, resulting in local trailing and blurred organ edges in the reconstructed images, affecting the accuracy of identifying fine needles and micro-bone sutures.

[0003] To address motion artifacts, recent solutions have mainly fallen into two categories: one is an inter-frame displacement compensation algorithm based on optical flow tracing, which predicts motion trajectories through the displacement vector fields of adjacent two frames of pixels; the other is a hardware synchronization between the detector and the robotic arm, which adjusts the exposure sequence according to a preset trajectory. The former relies on the assumption of motion continuity and generates incorrect compensation vectors for scenarios with sudden acceleration changes such as sudden stops of the equipment; the latter requires modification of the mechanical control system, increasing the complexity of the equipment and the risk of failure.

[0004] In current clinical practice, the above-mentioned approach still has three limitations: optical flow is prone to misinterpreting vascular texture as motion displacement in soft tissue deformation areas; hardware synchronization schemes cannot adapt to sudden changes in body position during surgery; existing deep learning deblurring networks excessively smooth bone crack textures below 0.5mm in pursuit of overall clarity; especially in minimally invasive spinal surgery, such defects can lead to surgeons misinterpreting the pin trajectory, requiring repeated imaging for verification and prolonging the operation time. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides an X-ray detector array imaging system that solves the problems of image blurring caused by non-uniform motion during C-arm surgery, the miscompensation of existing optical flow methods, and the loss of microvascular texture in deep learning deblurring.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides an X-ray detector array imaging system, comprising:

[0009] The image acquisition module is used to acquire multiple consecutive frames of X-ray projection images of the target object;

[0010] A motion compensation module, connected to the image acquisition module, is used to correct motion artifacts in the continuous multi-frame projection images;

[0011] The image reconstruction module is connected to the motion compensation module and outputs a three-dimensional reconstructed image.

[0012] The motion compensation module includes:

[0013] The input unit stacks three consecutive projected images into a three-dimensional tensor along the time dimension.

[0014] The acceleration estimation unit generates an acceleration channel based on the second derivative of the inter-frame displacement calculated from the three-dimensional tensor.

[0015] The dual-branch processing unit includes a parallel-connected spatial feature extraction branch and a temporal motion modeling branch.

[0016] As a preferred embodiment of the X-ray detector array imaging system of the present invention, the spatial feature extraction branch is composed of multiple residual convolutional blocks, which are used to extract anatomical structural features in the projected image.

[0017] The time motion modeling branch includes a Long Short-Term Memory (LSTM) network, with the input being the inter-frame displacement vector field calculated by optical flow.

[0018] As a preferred embodiment of the X-ray detector array imaging system of the present invention, the dual-branch processing unit further includes a fusion layer, which performs channel splicing of the output of the spatial feature extraction branch and the output of the time motion modeling branch.

[0019] The fusion layer connects to the adversarial training module, which includes a generator and a discriminator, wherein:

[0020] The generator outputs the corrected image;

[0021] The discriminator calculates adversarial loss based on the difference between the real high-resolution vascular texture image and the generated image.

[0022] In a preferred embodiment of the X-ray detector array imaging system of the present invention, the generation of the acceleration channel includes the following steps:

[0023] a) Calculate the pixel displacement vector field between two adjacent frames;

[0024] b) Perform time-dimension differentiation on the displacement vector field to obtain the acceleration field;

[0025] c) The acceleration field is superimposed onto the three-dimensional tensor as an independent channel.

[0026] As a preferred embodiment of the X-ray detector array imaging system of the present invention, wherein: in the acceleration channel generation step, the acceleration field is generated by performing second-order time-dimension differentiation, including:

[0027] At time index i, take the displacement vector field d of three adjacent frames. i-1 (x,y),d i (x,y),d i+1 (x,y) forms a local time window. Applying a Gaussian-weighted second-order central difference to this window yields a pixel-level acceleration vector:

[0028]

[0029] Among them, w +1 The time weight d represents the time weight when shifting forward one frame. i+1 (x, y) is the pixel displacement vector shifted forward one frame, in pixels (px), w0 is the time weight of the current frame, and d i (x, y) is the pixel displacement vector of the current frame, in pixels (px). -1 As the time weight for shifting backward by one frame, d i-1 (x, y) is the pixel displacement vector shifted one frame backward, in pixels (px), Δt is the sampling interval between two adjacent frames, in seconds (s), and a i (x, y) is a pixel-level acceleration vector, with units of px·s. -2 ;

[0030] The Gaussian kernel provides three weights:

[0031]

[0032] Among them, w r The normalized weights at a time offset of r frames, where r is the time offset in frames, and σ t This is the time spread coefficient, expressed in frames.

[0033] Adaptive suppression is introduced for the acceleration magnitude:

[0034]

[0035] A i (x,y)=‖a i (x,y)‖2,

[0036]

[0037] Among them, c i (x,y) represents the pixel-level suppression coefficient, k a px is the slope adjustment coefficient. -1 A i (x, y) represents the acceleration magnitude, in pixels per second (px·s). -2 γ a This is the acceleration threshold, measured in pixels per second (px·s). -2 , This represents the suppressed acceleration channel vector, in pixels (px·s). -2 .

[0038] As a preferred embodiment of the X-ray detector array imaging system of the present invention, the system is integrated into a C-arm X-ray machine, and the target object is the patient during surgery;

[0039] The acquisition angle of the continuous multi-frame projection images covers an arc trajectory from 0° to 180°.

[0040] In a preferred embodiment of the X-ray detector array imaging system described in this invention, the loss function comprises a weighted sum of the following three terms:

[0041] a) The sum of squared intensity differences between corresponding pixels in the corrected image output by the generator and the real high-resolution image;

[0042] b) The discriminator's binary cross-entropy loss for the difference in distribution between the generated image and the real image;

[0043] c) Perform a second difference operation along the time axis on the displacement vector field output by the time motion modeling branch, and calculate its Euclidean norm.

[0044] As a preferred embodiment of the X-ray detector array imaging system of the present invention, the loss function includes the calculation of the time quadratic difference and Euclidean norm of the displacement vector field, specifically comprising:

[0045] The time-motion modeling branch provides the pixel displacement vector field s in frame t. t (x,y); Perform a quadratic central difference on this sequence:

[0046]

[0047] Among them, s t+1 (x, y) is the pixel displacement vector at time index t+1, in pixels (px) and seconds (s). t (x, y) is the pixel displacement vector at time index t, in pixels (px) and seconds (s). t-1 (x, y) represents the pixel displacement vector at time index t-1, in pixels (px), Δt is the sampling interval between two adjacent frames, in seconds (s), and g t (x,y) is a second-order time difference vector of pixel-level displacement, in units of px·s. -2 ;

[0048] Taking the Euclidean norm of this difference vector yields the local curvature intensity:

[0049] r t (x,y)=‖g t (x,y)‖2,

[0050] Where, r t (x,y) represents the pixel-level curvature intensity, measured in pixels per second (px·s). -2 ;

[0051] We introduce a weight mask based on the acceleration channel to balance the contributions of different spatial regions in the loss function:

[0052]

[0053] Where ω(x,y) is the spatial weight, β is the weight amplification coefficient, α is the decay rate coefficient, and A(x,y) is the pixel-level acceleration modulus, in pixels per second (px·s). -2 A max This represents the maximum acceleration modulus of the current batch, in pixels per second (px·s). -2 ;

[0054] The combined time smoothing regularization term is obtained as follows:

[0055]

[0056] in, λ is the time smoothing regularization term. tem For regularization weights, N f N is the number of frames in the input sequence. p Total number of pixels in a single frame, in pixels (px).

[0057] The regularization weights are adaptively adjusted according to the overall motion intensity, and the adjustment formula is:

[0058]

[0059] Where η is the benchmark scaling factor. This is the average acceleration modulus, in pixels per second (px·s). -2 A ref This is a reference value for acceleration, in pixels (px·s). -2 .

[0060] As a preferred embodiment of the X-ray detector array imaging system of the present invention, when the image reconstruction module executes the filtering back projection algorithm, its projection data input is the image sequence processed by the motion compensation module, and the image sequence is weighted and filtered before reconstruction, and the filtering kernel function is negatively correlated with the amplitude of the acceleration channel.

[0061] As a preferred embodiment of the X-ray detector array imaging system of the present invention, the spatial feature extraction branch includes at least three residual convolutional blocks, each residual convolutional block being composed of a convolutional layer, a batch normalization layer and a ReLU activation layer connected in sequence;

[0062] The hidden layer dimension of the Long Short-Term Memory (LSTM) network in the temporal motion modeling branch is the same as the number of output channels in the spatial feature extraction branch.

[0063] The beneficial effects of this invention are as follows: The X-ray detector array imaging system provided by this invention significantly optimizes the dynamic imaging quality during surgery through multi-dimensional collaborative processing of the motion compensation module; based on the three-dimensional tensor input constructed from continuous frames, combined with the acceleration channel generation mechanism, the system can sensitively capture sudden displacement changes during surgery, such as non-uniform motion caused by sudden stopping of a chainsaw or respiratory muscle tremors. The curvature change of the displacement curve is accurately quantified through Gaussian weighted second-order difference, and weak jitter interference is filtered out using an adaptive suppression coefficient, providing highly robust time-varying characteristics for motion compensation.

[0064] The spatial feature extraction branch of the dual-branch processing unit of this invention preserves the anatomical details of bones and blood vessels, while the temporal motion modeling branch analyzes the long-term dependencies of the displacement field sequence through an LSTM network. The two branches are then concatenated by a fusion layer and input into an adversarial training module. Under the game mechanism of generator-discriminator, both overall motion artifacts are corrected, and microstructure loss caused by excessive smoothing is avoided. The curvature intensity penalty term for the displacement field in the loss function, combined with the acceleration-weighted spatial mask, effectively suppresses temporal jitter during motion compensation. This constraint is adaptively adjusted with the global acceleration level, and the smoothing restriction is automatically relaxed during fast-paced surgical movements to avoid the false suppression of real motion.

[0065] In the image reconstruction stage, a filter kernel with negative correlation of acceleration amplitude is introduced to further weaken the impact of residual artifacts on 3D reconstruction, so that the output image still maintains sharp organ edges in the scene of rapid instrument movement;

[0066] This invention achieves end-to-end optimization in continuous multi-angle imaging from 0° to 180° using a C-arm machine, reducing the need for repeated imaging of the same area during surgery and lowering the risk of radiation exposure for patients. Attached Figure Description

[0067] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 This is a schematic diagram of the framework of an X-ray detector array imaging system in Example 1. Detailed Implementation

[0069] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0070] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0072] Example 1, referring to Figure 1 This embodiment provides an X-ray detector array imaging system, including:

[0073] The image acquisition module is used to acquire multiple consecutive frames of X-ray projection images of the target object;

[0074] The motion compensation module, connected to the image acquisition module, is used to correct motion artifacts in multiple consecutive projected images;

[0075] The image reconstruction module connects to the motion compensation module and outputs a 3D reconstructed image.

[0076] The motion compensation module includes:

[0077] The input unit stacks three consecutive projected images into a three-dimensional tensor along the time dimension.

[0078] The acceleration estimation unit generates acceleration channels based on the second derivative of the inter-frame displacement calculated using a three-dimensional tensor.

[0079] The dual-branch processing unit includes a parallel-connected spatial feature extraction branch and a temporal motion modeling branch;

[0080] When the image reconstruction module executes the filtering back projection algorithm, its projection data input is the image sequence processed by the motion compensation module, and the image sequence is weighted and filtered before reconstruction. The filtering kernel function is negatively correlated with the amplitude of the acceleration channel.

[0081] The generation of acceleration channels includes the following steps:

[0082] a) Calculate the pixel displacement vector field between two adjacent frames;

[0083] b) Obtain the acceleration field by performing time-dimension differentiation on the displacement vector field;

[0084] c) Superimpose the acceleration field as an independent channel onto the three-dimensional tensor;

[0085] In the acceleration channel generation step, the acceleration field is generated by performing second-order time-dimension differentiation, including:

[0086] At time index i, take the displacement vector field d of three adjacent frames. i-1 (x,y),d i (x,y),d i+1 (x,y) forms a local time window. Applying a Gaussian-weighted second-order central difference to this window yields a pixel-level acceleration vector:

[0087]

[0088] Among them, w +1 The time weight d represents the time weight when shifting forward one frame. i+1 (x, y) is the pixel displacement vector shifted forward one frame, in pixels (px), w0 is the time weight of the current frame, and d i (x, y) is the pixel displacement vector of the current frame, in pixels (px). -1 As the time weight for shifting backward by one frame, d i-1 (x, y) is the pixel displacement vector shifted one frame backward, in pixels (px), Δt is the sampling interval between two adjacent frames, in seconds (s), and a i (x, y) is a pixel-level acceleration vector, with units of px·s. -2 ;

[0089] The Gaussian kernel provides three weights:

[0090]

[0091] Among them, w r The normalized weights at a time offset of r frames, where r is the time offset in frames, and σ t This is the time spread coefficient, expressed in frames.

[0092] To highlight rapid motion and suppress subtle jitter, adaptive suppression is introduced for the acceleration magnitude:

[0093]

[0094] A i (x,y)=‖a i (x,y)‖2,

[0095]

[0096] Among them, c i(x,y) represents the pixel-level suppression coefficient, k a px is the slope adjustment coefficient. -1 A i (x, y) represents the acceleration magnitude, in pixels per second (px·s). -2 γ a This is the acceleration threshold, measured in pixels per second (px·s). -2 , This represents the suppressed acceleration channel vector, in pixels (px·s). -2 ;

[0097] Specifically, this process constructs time-related weights using Gaussian kernels, making the difference more sensitive to short-term continuous motion and less affected by random jitter; the second-order central difference can directly reflect the curvature change of the displacement curve and locate the peak position of acceleration, thereby capturing the rapid movement of real tissue; the adaptive suppression stage uses a logistic function to smoothly prune the magnitude value, the signal in the weak motion region is low-pass filtered, and the original amplitude is maintained in the strong motion region, avoiding the injection of background noise into subsequent networks; the whole process forms a closed loop of weight allocation, curvature extraction, and amplitude modulation, and the acceleration channel is dynamically adjusted according to the rapid change of the shooting angle, providing high-contrast, high-robust time-varying prior features for motion compensation, significantly reducing artifact interference in angiography sequences;

[0098] The spatial feature extraction branch consists of multiple residual convolutional blocks, used to extract anatomical structural features from the projected image;

[0099] The temporal motion modeling branch includes a Long Short-Term Memory (LSTM) network, with the input being the inter-frame displacement vector field calculated by the optical flow method;

[0100] The dual-branch processing unit also includes a fusion layer, which concatenates the output of the spatial feature extraction branch with the output of the temporal motion modeling branch.

[0101] The fusion layer connects to the adversarial training module, which contains a generator and a discriminator, wherein:

[0102] The generator outputs the corrected image;

[0103] The discriminator calculates adversarial loss based on the difference between the real high-resolution vascular texture image and the generated image;

[0104] The spatial feature extraction branch contains at least three residual convolutional blocks, each of which consists of a convolutional layer, a batch normalization layer, and a ReLU activation layer connected in sequence.

[0105] The hidden layer dimension of the Long Short-Term Memory (LSTM) network in the temporal motion modeling branch is the same as the number of output channels in the spatial feature extraction branch;

[0106] The loss function consists of a weighted sum of the following three terms:

[0107] a) The sum of squared intensity differences between corresponding pixels in the corrected image output by the generator and the real high-resolution image;

[0108] b) The discriminator's binary cross-entropy loss for the difference in distribution between the generated image and the real image;

[0109] c) Perform a second difference operation on the displacement vector field output by the time motion modeling branch along the time axis and calculate its Euclidean norm;

[0110] The loss function involves calculating the time quadratic difference and Euclidean norm of the displacement vector field, specifically including:

[0111] The time-motion modeling branch provides the pixel displacement vector field s in frame t. t (x,y); Perform a quadratic central difference on this sequence:

[0112]

[0113] Among them, s t+1 (x, y) is the pixel displacement vector at time index t+1, in pixels (px) and seconds (s). t (x, y) is the pixel displacement vector at time index t, in pixels (px) and seconds (s). t-1 (x, y) represents the pixel displacement vector at time index t-1, in pixels (px), Δt is the sampling interval between two adjacent frames, in seconds (s), and g t (x,y) is a second-order time difference vector of pixel-level displacement, in units of px·s. -2 ;

[0114] Taking the Euclidean norm of this difference vector yields the local curvature intensity:

[0115] r t (x,y)=‖g t (x,y)‖2,

[0116] Where, r t (x,y) represents the pixel-level curvature intensity, measured in pixels per second (px·s). -2 ;

[0117] We introduce a weight mask based on the acceleration channel to balance the contributions of different spatial regions in the loss function:

[0118]

[0119] Where ω(x,y) is the spatial weight, β is the weight amplification coefficient, α is the decay rate coefficient, and A(x,y) is the pixel-level acceleration modulus, in pixels per second (px·s). -2 A max This represents the maximum acceleration modulus of the current batch, in pixels per second (px·s). -2;

[0120] The combined time smoothing regularization term is obtained as follows:

[0121]

[0122] in, λ is the time smoothing regularization term. tem For regularization weights, N f N is the number of frames in the input sequence. p Total number of pixels in a single frame, in pixels (px).

[0123] The regularization weights are adaptively adjusted according to the overall motion intensity, and the adjustment formula is:

[0124]

[0125] Where η is the benchmark scaling factor. This is the average acceleration modulus, in pixels per second (px·s). -2 A ref This is a reference value for acceleration, in pixels (px·s). -2 ;

[0126] Specifically, the quadratic central difference directly reflects the curvature of the displacement sequence and can mark the location of acceleration mutations; after the Euclidean norm is converted into a scalar, the curvature intensity of each pixel in the same frame is easy to measure uniformly; the weight mask allows regions with high acceleration amplitude to occupy a larger proportion in the regularization term, while low-amplitude static regions have limited impact on training, thus highlighting real anatomical motion; the regularization weights rely on the adaptive change of global average acceleration, automatically relaxing the smoothing constraint when fast-paced movements occur during surgery to avoid over-suppression, and increasing the constraint when movements are slow to suppress temporal noise; this curvature extraction-spatial weighting-global adaptive chain improves the sensitivity of the motion modeling branch to temporal consistency, while reducing the network's tendency to overfit pseudo-random transitions;

[0127] This system is integrated into a C-arm X-ray machine and is designed for patients during surgery.

[0128] The acquisition angle of the continuous multi-frame projection images covers an arc trajectory from 0° to 180°.

[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An X-ray detector array imaging system, characterized in that, include, The image acquisition module is used to acquire multiple consecutive frames of X-ray projection images of the target object; A motion compensation module, connected to the image acquisition module, is used to correct motion artifacts in the continuous multi-frame projection images; The image reconstruction module is connected to the motion compensation module and outputs a three-dimensional reconstructed image. The motion compensation module includes: The input unit stacks three consecutive projected images into a three-dimensional tensor along the time dimension. The acceleration estimation unit generates an acceleration channel based on the second derivative of the inter-frame displacement calculated from the three-dimensional tensor. The dual-branch processing unit includes a parallel-connected spatial feature extraction branch and a temporal motion modeling branch.

2. The X-ray detector array imaging system as described in claim 1, characterized in that, The spatial feature extraction branch consists of multiple residual convolutional blocks, used to extract anatomical structural features from the projected image; The time motion modeling branch includes a Long Short-Term Memory (LSTM) network, with the input being the inter-frame displacement vector field calculated by optical flow.

3. The X-ray detector array imaging system as described in claim 2, characterized in that, The dual-branch processing unit also includes a fusion layer, which performs channel splicing between the output of the spatial feature extraction branch and the output of the temporal motion modeling branch. The fusion layer connects to the adversarial training module, which includes a generator and a discriminator, wherein: The generator outputs the corrected image; The discriminator calculates adversarial loss based on the difference between the real high-resolution vascular texture image and the generated image.

4. The X-ray detector array imaging system as described in claim 1, characterized in that, The generation of the acceleration channel includes the following steps: a) Calculate the pixel displacement vector field between two adjacent frames; b) Perform time-dimension differentiation on the displacement vector field to obtain the acceleration field; c) The acceleration field is superimposed onto the three-dimensional tensor as an independent channel.

5. The X-ray detector array imaging system as described in claim 4, characterized in that, In the acceleration channel generation step, the acceleration field is generated by performing second-order time-dimension differentiation, including: At time index i, take the displacement vector field d of three adjacent frames. i-1 (x,y),d i (x,y),d i+1 (x,y) forms a local time window. Applying a Gaussian-weighted second-order central difference to this window yields a pixel-level acceleration vector: Among them, w +1 The time weight d represents the time weight when shifting forward one frame. i+1 (x, y) is the pixel displacement vector shifted forward one frame, in pixels (px), w0 is the time weight of the current frame, and d i (x, y) is the pixel displacement vector of the current frame, in pixels (px). -1 As the time weight for shifting backward by one frame, d i-1 (x, y) is the pixel displacement vector shifted one frame backward, in pixels (px), Δt is the sampling interval between two adjacent frames, in seconds (s), and a i (x, y) is a pixel-level acceleration vector, with units of px·s. -2 ; The Gaussian kernel provides three weights: Among them, w r The normalized weights at a time offset of r frames, where r is the time offset in frames, and σ t This is the time spread coefficient, expressed in frames. Adaptive suppression is introduced for the acceleration magnitude: TO i (x,y)=‖a i (x,y)‖2, Among them, c i (x,y) represents the pixel-level suppression coefficient, k a px is the slope adjustment coefficient. -1 A i (x, y) represents the acceleration magnitude, in pixels per second (px·s). -2 γ a This is the acceleration threshold, measured in pixels per second (px·s). -2 , This represents the suppressed acceleration channel vector, in pixels (px·s). -2 .

6. An X-ray detector array imaging system as described in any one of claims 1 to 5, characterized in that, The system is integrated into a C-arm X-ray machine, and the target is the patient during surgery. The acquisition angle of the continuous multi-frame projection images covers an arc trajectory from 0° to 180°.

7. The X-ray detector array imaging system as described in claim 3, characterized in that, The loss function consists of a weighted sum of the following three terms: a) The sum of squared intensity differences between corresponding pixels in the corrected image output by the generator and the real high-resolution image; b) The discriminator's binary cross-entropy loss for the difference in distribution between the generated image and the real image; c) Perform a second difference operation along the time axis on the displacement vector field output by the time motion modeling branch, and calculate its Euclidean norm.

8. The X-ray detector array imaging system as described in claim 7, characterized in that, The loss function includes the calculation of the time quadratic difference and Euclidean norm of the displacement vector field, specifically: The time-motion modeling branch provides the pixel displacement vector field s in frame t. t (x,y); Perform a quadratic central difference on this sequence: Among them, s t+1 (x, y) is the pixel displacement vector at time index t+1, in pixels (px) and seconds (s). t (x, y) is the pixel displacement vector at time index t, in pixels (px) and seconds (s). t-1 (x, y) represents the pixel displacement vector at time index t-1, in pixels (px), Δt is the sampling interval between two adjacent frames, in seconds (s), and g t (x,y) is a second-order time difference vector of pixel-level displacement, in units of px·s. -2 ; Taking the Euclidean norm of this difference vector yields the local curvature intensity: r t (x,y)=‖g t (x,y)‖2, Where, r t (x,y) represents the pixel-level curvature intensity, measured in pixels per second (px·s). -2 ; We introduce a weight mask based on the acceleration channel to balance the contributions of different spatial regions in the loss function: Where ω(x,y) is the spatial weight, β is the weight amplification coefficient, α is the decay rate coefficient, and A(x,y) is the pixel-level acceleration modulus, in pixels per second (px·s). -2 A max This represents the maximum acceleration modulus of the current batch, in pixels per second (px·s). -2 ; The time smoothing regularization term is obtained by synthesis: in, λ is the time smoothing regularization term. tem For regularization weights, N f N is the number of frames in the input sequence. p Total number of pixels in a single frame, in pixels (px). The regularization weights are adaptively adjusted according to the overall motion intensity, and the adjustment formula is: Where η is the benchmark scaling factor. This is the average acceleration modulus, in pixels per second (px·s). -2 A ref This is a reference value for acceleration, in pixels (px·s). -2 .

9. An X-ray detector array imaging system as described in claim 1, characterized in that, When the image reconstruction module executes the filtered back projection algorithm, its projection data input is the image sequence processed by the motion compensation module, and the image sequence is weighted and filtered before reconstruction. The filtering kernel function is negatively correlated with the amplitude of the acceleration channel.

10. An X-ray detector array imaging system as described in claim 2, characterized in that, The spatial feature extraction branch contains at least three residual convolutional blocks, each of which is composed of a convolutional layer, a batch normalization layer, and a ReLU activation layer connected in sequence. The hidden layer dimension of the Long Short-Term Memory (LSTM) network in the temporal motion modeling branch is the same as the number of output channels in the spatial feature extraction branch.

Images