A portable DR image intelligent enhancement method and system

Abstract

The application relates to the technical field of DR image processing, and discloses a portable DR image intelligent enhancement method and system. The method obtains vibration information generated by a portable DR detector in an X-ray exposure imaging process, adjusts an X-ray exposure mode to a multi-frame exposure mode according to the vibration information, and collects multiple original X-ray images. The method further calculates relative motion vectors between the multiple original X-ray images according to pixel gray matching results between the multiple original X-ray images, carries out motion compensation alignment processing on the original images by using the relative motion vectors, effectively eliminates frame position offset errors, obtains multiple aligned images, carries out pixel-level weighted fusion operation on the multiple aligned images on the basis, and generates a fusion image, so that image motion blurring caused by the vibration information is significantly inhibited.

Description

Technical Field

[0001] This invention relates to the field of DR image processing technology, and more specifically, to a portable DR image intelligent enhancement method and system. Background Technology

[0002] In the field of emergency medical care, portable DR devices play an irreplaceable role in complex environments such as wilderness or disaster sites due to their lightweight and rapid imaging capabilities. However, in complex environments, such as wilderness rescue or low-light conditions, the images acquired by these devices are often affected by noise interference and insufficient contrast, making it difficult for doctors to see subtle lesions. Existing image enhancement technologies often rely on adjusting a few fixed parameters, making it difficult to flexibly adapt to constantly changing clinical needs.

[0003] In practical applications, such as emergency rescue sites during geological disasters in mountainous areas, medical rescue teams use portable DR equipment for initial screening of the injured. The on-site environment is complex, with dim lighting, and continuous low-frequency vibrations on the ground due to the continuous operation of large rescue machinery. During X-ray exposure imaging, this vibration may cause slight relative displacement between the X-ray tube or detector and the injured person, resulting in motion blur in the original X-ray image, characterized by directional trailing at the edges and details of the image.

[0004] The intelligent enhancement systems built into existing portable DR devices typically execute preset image quality enhancement modules after acquiring raw X-ray images. These modules include noise reduction and contrast enhancement procedures. However, these general-purpose modules are not specifically trained to recognize and process motion blur. When the trailing shadows caused by motion blur are incorrectly identified as details requiring sharpening, the sharpening component over-enhances these shadow edges, causing originally blurred edges to become abnormally sharp and even creating subtle linear or mesh-like artifacts in the image.

[0005] Furthermore, when the system activates an optimized display scheme for specific anatomical locations (such as the pleural cavity), these linear artifacts, generated by motion blur and false sharpening, may, after further enhancement, have a shape and location that visually resemble the lines of free gas within the pleural cavity during pneumothorax, thus producing clinically misleading artifacts. These artifacts could lead frontline medical staff to make incorrect treatment decisions based on erroneous information, causing secondary harm to patients. Therefore, in complex application scenarios, portable DR image intelligent enhancement methods urgently need to address the problem of accurately identifying and processing motion blur to avoid generating clinically misleading artifacts.

[0006] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0007] This invention provides a portable DR image intelligent enhancement method and system, aiming to solve the technical problem that portable DR devices are easily affected by vibration in complex environments, resulting in motion blur of images. Furthermore, existing enhancement technologies may misidentify motion blur as details and over-sharpen them, or even produce clinically misleading artifacts, thereby affecting the accuracy of diagnosis.

[0008] The technical solution of this application is as follows: In a first aspect, this application discloses a portable DR image intelligent enhancement method, which includes: Acquire vibration information generated by a portable DR detector during X-ray exposure imaging; Based on the vibration information, the X-ray exposure mode was adjusted to a multi-frame exposure mode, and multiple raw X-ray images were acquired. Based on the pixel grayscale matching results between multiple original X-ray images, the relative motion vector between each frame image is calculated; Based on the relative motion vector, multiple original X-ray images are subjected to motion compensation and alignment processing to eliminate inter-frame position offset errors and obtain multiple aligned images. A pixel-level weighted fusion operation is performed on multiple aligned images to generate a fused image, thereby suppressing motion blur caused by vibration information. Local texture features are extracted from the fused image. The local texture features are compared with the preset range of normal anatomical structure features to identify abnormal areas where the features deviate from the preset range, and the abnormal areas are determined to be real anatomical structures or image artifacts. If the abnormal region is determined to be an image artifact, the weight coefficients corresponding to the pixel-level weighted fusion are adjusted, the image fusion operation is re-executed, and an optimized fused image is generated.

[0009] This technical solution effectively addresses the image motion blur problem caused by vibration in portable DR devices under complex environments. Through multi-frame exposure, motion compensation, and intelligent fusion, it suppresses blur while further avoiding clinically misleading artifacts that may occur with traditional methods by using artifact recognition and fusion weight adjustment mechanisms, thus significantly improving image quality and diagnostic accuracy.

[0010] Furthermore, when calculating the relative motion vectors between each frame, the method includes: One frame is selected from multiple original X-ray images as a reference frame, and multiple pixels with gray-scale gradient changes are extracted from the reference frame as feature tracking points. In the remaining subsequent frames, pixel coordinates of each feature tracking point are tracked one by one to obtain the displacement vector corresponding to each tracking point. By integrating the displacement vector data of all feature tracking points, the global translation vector and global rotation angle between the reference frame and each subsequent frame are obtained through vector fitting, thus obtaining the inter-frame relative motion vector.

[0011] Through this technical solution, this application can accurately quantify the global translation and rotation between frames by precise feature tracking and vector fitting, providing a reliable basis for subsequent motion compensation, thereby more effectively eliminating the overall motion offset of the image.

[0012] Based on this, when performing motion compensation alignment on multiple original X-ray images, the method includes: Based on the global translation vector and the global rotation angle, construct the pixel coordinate transformation mapping relationship between each subsequent frame image and the reference frame; Perform pixel-level spatial transformation on each subsequent frame image according to the coordinate transformation mapping relationship, so that the anatomical structure pixel position of each subsequent frame is aligned with the corresponding anatomical structure pixel position of the reference frame. Interpolation is performed on the missing pixel positions in the spatially transformed image to eliminate transformation holes and obtain multiple complete aligned images.

[0013] This technical solution enables precise alignment of multiple frames of images by constructing accurate coordinate transformation mapping relationships and performing pixel-level spatial transformations, effectively eliminating inter-frame position offset errors and ensuring the image quality of subsequent fusion.

[0014] In some preferred embodiments, the pixel-level weighted fusion operation on multiple aligned images specifically includes: Calculate the local gray-level statistical features of each pixel position in each aligned image pixel by pixel, and quantize to obtain the noise level estimate of the corresponding pixel position; The fusion weights are adaptively assigned based on the noise level estimate of each pixel location, with higher fusion weights assigned to pixel locations with lower noise level estimates. The gray values ​​at the same pixel position in multiple aligned images are weighted and summed, and the weighted sum is used as the gray value of the corresponding pixel position in the fused image, thus completing pixel-level image fusion.

[0015] This technical solution effectively suppresses noise in an image by adaptively allocating fusion weights based on pixel noise levels, ensuring that high-quality pixels dominate the fusion process, thereby generating a clearer, less noisy fused image.

[0016] Furthermore, when distinguishing between abnormal regions as genuine anatomical structures and image artifacts, the method includes: Gradient information of edge pixels in abnormal regions is extracted, and the gradient direction distribution characteristics of edge pixels are statistically obtained. If the gradient direction of the edge pixels is uniformly distributed in multiple directions, then the abnormal area is determined to be an image artifact. If the gradient direction of edge pixels is concentrated in a single dominant direction, then the abnormal region is determined to be a real anatomical structure.

[0017] By using this technical solution, this application can accurately distinguish between artifacts caused by motion blur or noise and real anatomical structures by analyzing the gradient direction distribution characteristics of edge pixels in abnormal regions, thus avoiding misjudgment and over-enhancement of artifacts.

[0018] Based on the above, when adjusting the weight coefficients corresponding to pixel-level weighted fusion and re-performing the image fusion operation to generate an optimized fused image, the method includes: Locate abnormal regions that are image artifacts in the fused image, and extract the pixel positions of the corresponding regions to construct a set of artifact pixel positions; Reduce the fusion weight of pixel positions corresponding to the artifact pixel position set in multiple aligned images; Pixel-level weighted fusion is re-executed based on the adjusted fusion weights to generate an optimized fused image with artifacts removed.

[0019] By using this technical solution, this application can effectively suppress the generation and enhancement of artifacts by specifically reducing the fusion weight of artifact regions, thereby generating a cleaner and more accurate optimized fusion image.

[0020] As a technical improvement, the method also includes an iterative optimization step, which includes: Regenerate the optimized fused image and extract local texture features from the new fused image; The re-extracted local texture features are compared a second time with the preset range of normal anatomical structure features; If no new abnormal regions are identified in the second comparison, the current fused image is output as the final enhanced image; If abnormal regions are still identified in the second comparison, the steps of adjusting the weight coefficients and regenerating the fused image are repeated until there are no abnormal regions.

[0021] Through this technical solution, this application can continuously detect and eliminate artifacts in the image through an iterative optimization mechanism until the image quality reaches an optimal state, ensuring that the final output enhanced image does not contain misleading artifacts.

[0022] As a further improvement, the method also includes a fusion weight optimization step based on vibration time period, which includes: When acquiring multiple raw X-ray images, the acquisition timestamps of each frame are recorded simultaneously. Determine the peak periods when the vibration amplitude exceeds a preset amplitude threshold based on vibration information; In the pixel-level weighted fusion process, the overall fusion weight of image frames whose acquisition timestamps fall within the peak period is reduced to weaken the interference of high-vibration, low-quality images on the fusion result.

[0023] By using this technical solution, this application can effectively avoid the negative impact of inferior images on the final fusion result by identifying high vibration periods and reducing the fusion weight of the corresponding image frames, thereby further improving image quality.

[0024] As a system extension, the method also includes an exposure parameter adaptive optimization step, which includes: Obtain the current exposure dose parameters of the portable DR detector; Based on the vibration amplitude value of the vibration information, the exposure dose adjustment factor is determined according to the preset mapping relationship between amplitude and dose. In this mapping relationship, the vibration amplitude value and the exposure dose adjustment factor are monotonically increasing. Multiply the current exposure dose parameter by the exposure dose adjustment factor to obtain the optimized exposure dose parameter, which is used to optimize the quality of the original image acquisition.

[0025] This technical solution enables the application to optimize the acquisition quality of the original image under vibration by adaptively adjusting the exposure dose according to the vibration amplitude, thereby improving the image data from the source and providing a better foundation for subsequent image enhancement processing.

[0026] Secondly, this application also discloses a portable DR image intelligent enhancement system for performing the above-mentioned portable DR image intelligent enhancement method. The system includes: a vibration monitoring module, an exposure control module, a motion processing module, an image fusion module, an artifact detection module, an artifact discrimination module, and a fusion adjustment module. The vibration monitoring module is used to acquire vibration information generated by the portable DR detector during X-ray exposure imaging. The exposure control module is used to adjust the X-ray exposure mode to a multi-frame exposure mode based on vibration information and acquire multiple raw X-ray images. The motion processing module is used to calculate the inter-frame relative motion vector based on the pixel grayscale matching results of multiple original X-ray images, and to perform motion compensation and alignment processing on the original images based on the relative motion vector to obtain multiple aligned images. The image fusion module is used to perform pixel-level weighted fusion of multiple aligned images to generate a fused image and suppress motion blur caused by vibration. The artifact detection module is used to extract local texture features from the fused image, compare the local texture features with a preset range of normal anatomical structure features, and identify abnormal areas in the image. The artifact discrimination module is used to distinguish between real anatomical structures and image artifacts based on the edge gradient distribution characteristics of abnormal regions. The fusion adjustment module is used to adjust the weight coefficients of pixel-level weighted fusion when image artifacts are detected, and regenerate the optimized fused image to achieve image enhancement optimization.

[0027] This application provides a system that can effectively implement a portable DR image intelligent enhancement method. Through modular design, it realizes functions such as vibration monitoring, multi-frame exposure, motion compensation, intelligent fusion, artifact detection and discrimination, and fusion adjustment, thereby supporting a significant improvement in image quality and effective avoidance of clinically misleading artifacts at the hardware level.

[0028] The portable DR image intelligent enhancement method and system disclosed in this application acquires vibration information generated by the portable DR detector during X-ray exposure imaging, and adjusts the X-ray exposure mode to a multi-frame exposure mode based on this vibration information to acquire multiple original X-ray images. The method further calculates the relative motion vector between each frame image based on the pixel grayscale matching results between the multiple original X-ray images, and uses this relative motion vector to perform motion compensation alignment processing on the original images, effectively eliminating inter-frame position offset errors and obtaining multiple aligned images. Based on this, pixel-level weighted fusion operations are performed on the multiple aligned images to generate a fused image, thereby significantly suppressing image motion blur caused by vibration information.

[0029] More importantly, this method extracts local texture features from the fused image and compares them with a preset range of normal anatomical structure features to identify abnormal regions where the features deviate from the preset range. It then intelligently determines whether the abnormal region is a real anatomical structure or an image artifact. If it is determined to be an image artifact, the weight coefficients corresponding to the pixel-level weighted fusion are adjusted, and the image fusion operation is re-executed to generate an optimized fused image.

[0030] Through the above technical solution, this application effectively solves the problems in the prior art where portable DR devices cause image motion blur due to vibration in complex environments, and where existing image enhancement techniques may misidentify motion blur as detail and over-sharpen it, even producing clinically misleading artifacts. This method suppresses motion blur at its source through multi-frame exposure and motion compensation, and avoids the generation and enhancement of artifacts through intelligent artifact recognition and fusion weight adjustment mechanisms, especially avoiding clinically misleading artifacts that may be confused with real lesions. Therefore, this application can significantly improve the quality and diagnostic accuracy of portable DR images, providing more reliable image diagnostic evidence for complex application scenarios such as medical emergency care, and has significant clinical application value. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating a portable DR image intelligent enhancement method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a process for calculating the relative motion vectors between frames of images according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a portable DR image intelligent enhancement system provided in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Reference Figure 1 , Figure 1 This is a flowchart illustrating a portable DR image intelligent enhancement method provided by an embodiment of the present invention, including the following steps: S11, acquire vibration information generated by the portable DR detector during X-ray exposure imaging; S12, Based on the vibration information, the X-ray exposure mode is adjusted to a multi-frame exposure mode, and multiple original X-ray images are acquired; S13, calculate the relative motion vector between each frame image based on the pixel grayscale matching results between the multiple original X-ray images; S14, Based on the relative motion vector, perform motion compensation and alignment processing on the multiple original X-ray images to eliminate inter-frame position offset errors and obtain multiple aligned images; S15, perform pixel-level weighted fusion operation on the multiple aligned images to generate a fused image and suppress image motion blur caused by vibration information; S16, extract the local texture features of the fused image, compare the local texture features with the preset range of normal anatomical structure features, identify abnormal areas where the features deviate from the preset range, and determine whether the abnormal areas are real anatomical structures or image artifacts. S17, if the abnormal region is determined to be an image artifact, the weight coefficients corresponding to the pixel-level weighted fusion are adjusted, the image fusion operation is re-executed, and an optimized fused image is generated.

[0034] This application effectively suppresses motion blur caused by vibration by acquiring vibration information and adjusting the exposure mode to multi-frame exposure, combined with motion compensation alignment and pixel-level weighted fusion. Simultaneously, an artifact recognition and discrimination mechanism is introduced, and the fusion weights are adjusted based on the discrimination results, achieving intelligent removal of image artifacts. This generates high-quality optimized fused images, avoiding clinically misleading artifacts and significantly improving the diagnostic accuracy of portable DR images.

[0035] To better understand the technical solutions proposed in this application, it is necessary to explain some key terms and implementation environments involved. A portable DR detector is a lightweight, mobile digital X-ray imaging device, commonly used in field, emergency, or bedside settings. The X-ray exposure imaging process refers to the process by which X-rays penetrate the object being examined, are received by the detector, and converted into a digital image. Vibration information refers to the data on minute mechanical movements of the detector or the object being examined during the imaging process due to external environmental factors or the equipment itself, such as displacement, velocity, or acceleration data obtained through an accelerometer or gyroscope. Multi-frame exposure mode refers to the detector continuously acquiring multiple images under a single X-ray irradiation, rather than a single image. The raw X-ray image is the unprocessed digital image directly acquired by the detector. Pixel grayscale matching result refers to comparing the similarity or difference of pixel grayscale values ​​between different image frames using image processing algorithms. Relative motion vector refers to a mathematical vector describing the relative translation and rotation between different image frames.

[0036] Motion-compensated alignment refers to performing geometric transformations on images based on relative motion vectors to align them spatially. Pixel-level weighted fusion refers to averaging the same pixel positions in multiple aligned images according to preset or adaptive weights to generate a fused image. Local texture features refer to visual characteristics such as grayscale variations, edge directions, and texture density in local areas of an image. The preset range of normal anatomical structure features refers to the parameter range used to describe the texture features of normal human anatomical structures, set through training with a large number of normal DR images or expert experience. Abnormal regions refer to areas in an image whose local texture features deviate from the range of normal anatomical structure features. Image artifacts refer to image distortions or artifacts introduced by the imaging system or processing that do not belong to the true anatomical structure.

[0037] The portable DR image intelligent enhancement method proposed in this application is based on a series of refined image processing steps to effectively address the challenges of portable DR imaging in complex environments.

[0038] Firstly, various methods can be employed to acquire vibration information generated by portable DR detectors during X-ray exposure imaging. For example, a high-precision accelerometer or gyroscope can be integrated inside the portable DR detector to monitor minute vibrations during exposure in real time. These sensors can continuously acquire acceleration or angular velocity data of the detector in the X, Y, and Z directions at a high sampling rate, and convert this data into quantitative indicators such as vibration amplitude and frequency through a data processing unit. As a preferred implementation, an independent vibration monitoring device, such as a laser vibrometer, can be installed outside the detector to measure the vibration displacement of the detector surface in a non-contact manner and transmit the measurement data to the image processing system. This vibration information provides crucial information for subsequent exposure mode adjustments and image processing.

[0039] Secondly, based on the vibration information, the X-ray exposure mode is adjusted to a multi-frame exposure mode, and multiple raw X-ray images are acquired. When the system detects that the vibration amplitude exceeds a preset threshold, the multi-frame exposure mode can be automatically triggered. For example, in single-frame exposure mode, the system may only acquire one image; if vibration is present at this time, the image will directly contain motion blur. In multi-frame exposure mode, the system continuously acquires multiple images at extremely short intervals within one X-ray irradiation cycle, such as 5 frames, 10 frames, or even more. Although these images may all be affected by vibration, because vibration is dynamic, the degree and direction of blur in different frames may differ, providing redundant information for subsequent motion compensation and fusion.

[0040] Next, based on the pixel grayscale matching results between the multiple original X-ray images, the relative motion vector between each frame is calculated. After acquiring multiple original X-ray images, it is necessary to determine their relative motion relationships. One approach is to extract unique feature points in adjacent frames using feature-point matching algorithms, such as SIFT (Scale Invariant Feature Transform) or SURF (Accelerated Robust Feature Transform), and calculate the correspondence between these feature points across different frames. The translational and rotational motions between frames can be estimated using the displacements of these corresponding feature points. As a specific implementation, a block-based matching algorithm is used to divide a frame into multiple small blocks, and then search for the region most similar to these small blocks in another frame, thereby determining the displacement of each small block. Statistical analysis of the displacements of all small blocks yields the global relative motion vector.

[0041] Then, based on the relative motion vectors, motion compensation alignment is performed on the multiple original X-ray images to eliminate inter-frame positional offset errors, resulting in multiple aligned images. Once the relative motion vectors between each frame are calculated, these vectors can be used to perform geometric transformations on the images to align them spatially. For example, if a frame is detected to have a rightward translation and clockwise rotation relative to a reference frame, the anatomical structure pixel positions of that frame can be adjusted to correspond to the positions in the reference frame through image translation and rotation transformations. This alignment process effectively eliminates inter-frame positional offsets caused by vibration, laying the foundation for subsequent image fusion.

[0042] Subsequently, pixel-level weighted fusion is performed on the multiple aligned images to generate a fused image, suppressing motion blur caused by vibration information. After obtaining multiple aligned images, they need to be fused into a single high-quality image. A simple fusion method is to take the arithmetic mean of the same pixel positions in all aligned images. However, this method may not adequately suppress noise and blur. A better fusion method is to use pixel-level weighted fusion. For example, different weights can be assigned to each pixel based on the local sharpness or noise level of each aligned image. Pixels with high sharpness or low noise levels are given higher weights, and vice versa. Through this weighted summation, a fused image that suppresses motion blur and reduces noise can be generated.

[0043] Furthermore, local texture features of the fused image are extracted and compared with a preset range of normal anatomical structure features to identify abnormal regions where features deviate from the preset range, and to determine whether the abnormal regions are real anatomical structures or image artifacts. After generating the fused image, its quality needs to be evaluated, especially to identify the presence of artifacts. Image processing algorithms, such as Gabor filters or Local Binary Patterns (LBP), can be used to extract texture features of various local regions in the fused image, such as edge direction, texture thickness, and contrast. These extracted texture features are compared with a pre-established database of normal anatomical structure features. For example, normal bone texture has specific directionality and density, while artifacts may appear as irregular lines or spots. When the texture features of a certain region significantly deviate from the normal range, that region is marked as an abnormal region.

[0044] Finally, if the abnormal region is determined to be an image artifact, the weight coefficients corresponding to the pixel-level weighted fusion are adjusted, and the image fusion operation is re-executed to generate an optimized fused image. If the abnormal region is identified as an image artifact, the image fusion process needs to be optimized. For example, the image region where the artifact is located can be located, and when re-executing pixel-level weighted fusion, the weights of those original image frames that contribute significantly to the artifact region can be reduced, or the pixels in the artifact region can be specially processed, such as using a smoother fusion strategy. Through this iterative optimization, a higher-quality optimized fused image with artifacts removed can be generated.

[0045] The portable DR image intelligent enhancement method proposed in this application effectively solves the problem of image blurring and artifacts caused by vibration in traditional portable DR imaging under complex environments through a series of synergistic technical features.

[0046] In summary, this application establishes a closed-loop intelligent enhancement system through vibration-driven multi-frame exposure, precise motion compensation alignment, intelligent artifact recognition and discrimination, and adaptive fusion optimization. This not only significantly improves the clarity and detail of portable DR images in complex environments, but more importantly, it effectively avoids clinically misleading artifacts, thereby greatly improving the accuracy and safety of diagnosis. It provides more reliable image diagnostic support for scenarios such as medical emergency care, demonstrating significant progressiveness and practical value.

[0047] For details, please refer to Figure 2 , Figure 2 This is a flowchart illustrating a method for calculating the relative motion vectors between frames of an image according to an embodiment of the present invention, including: S131, Select one frame from the multiple original X-ray images as a reference frame, and extract multiple pixel points with gray-scale gradient changes as feature tracking points from the reference frame. S132, in the remaining subsequent frame images, pixel coordinate tracking is performed on each feature tracking point to obtain the displacement vector corresponding to each tracking point; S133, integrate the displacement vector data of all feature tracking points, and obtain the global translation vector and global rotation angle between the reference frame and each subsequent frame through vector fitting, thereby obtaining the inter-frame relative motion vector.

[0048] As a preferred implementation, a frame with the best image quality, least vibration impact, or most timely position can be selected as a reference frame so that subsequent motion tracking and alignment operations can be performed on a stable benchmark. Pixels with grayscale gradient changes typically correspond to edges, corners, or textured regions in the image. These regions have high distinguishability and stability in the image sequence and are more suitable as feature points for tracking.

[0049] For example, algorithms such as Harris corner detection, SIFT (Scale Invariant Feature Transform), or SURF (Speed-Up Robust Features) can be used to automatically identify and extract these feature points. Pixel coordinate tracking refers to identifying and locating the position of the same feature point in different frames within consecutive image frames. This can be achieved through optical flow methods (such as Lucas-Kanade optical flow), block matching algorithms, or feature descriptor-based matching methods. The goal of tracking is to obtain the precise displacement information of each feature point from the reference frame to subsequent frames. Each displacement vector represents the pixel-level movement direction and distance of a feature point between two image frames. After integrating the displacement vector data of all feature points, robust vector fitting methods (e.g., RANSAC (Random Sample Consensus) algorithm combined with least squares) are needed to filter out outliers (such as those caused by local deformation or tracking errors) and estimate the global motion parameters of the entire image. The global translation vector describes the amount of translation of the entire image in the X and Y directions, while the global rotation angle describes the amount of rotation of the entire image. The resulting global translation vector and global rotation angle together constitute the inter-frame relative motion vector, which accurately quantifies the overall motion relationship between adjacent or reference frames and subsequent frames.

[0050] The proposed method first establishes a stable reference frame and then extracts highly discriminative feature tracking points based on this frame, ensuring the reliability of motion tracking. Subsequently, by tracking the pixel coordinates of these feature points frame by frame, the motion information of various local regions in the image can be accurately captured. Furthermore, by integrating the displacement vector data of all tracking points and performing robust vector fitting, the effects of local noise and non-rigid deformation can be effectively filtered out, thereby accurately solving for the global translation vector and global rotation angle of the entire image. This method, based on feature point tracking and global motion fitting, overcomes the shortcomings of traditional whole-image matching methods that are sensitive to local deformation and noise, providing high-precision motion parameters for subsequent motion compensation and alignment processing.

[0051] Specifically, the motion compensation and alignment process for multiple original X-ray images described above includes the following steps: Based on the global translation vector and global rotation angle mentioned above, construct the pixel coordinate transformation mapping relationship between each subsequent frame image and the reference frame; According to the coordinate transformation mapping relationship, perform pixel-level spatial transformation on each subsequent frame image so that the anatomical structure pixel position of each subsequent frame is aligned with the corresponding anatomical structure pixel position of the reference frame. Interpolation is performed on the missing pixel positions in the spatially transformed image to eliminate transformation holes and obtain multiple complete aligned images.

[0052] Specifically, after obtaining the global translation vector and global rotation angle between the reference frame and each subsequent frame, these parameters can be used to construct a two-dimensional affine transformation matrix or a more complex geometric transformation model. This model defines how each pixel in the subsequent frame is mapped to the coordinate system of the reference frame, thus forming a pixel coordinate transformation mapping relationship. For example, for the position of a pixel in a subsequent frame, its corresponding position in the reference frame can be calculated using this mapping relationship.

[0053] Pixel-level spatial transformation refers to resampling the position of each pixel in subsequent frames of the image according to the established coordinate transformation mapping. This typically involves moving the pixels of subsequent frames to new coordinate positions according to the mapping relationship to achieve alignment with the anatomical pixel positions of the reference frame. For example, nearest neighbor interpolation, bilinear interpolation, or bicubic interpolation can be used to determine the pixel grayscale value at the new position.

[0054] In practical applications, after performing pixel-level spatial transformations, due to the discreteness of pixels and the nature of the transformation, some missing regions without corresponding pixel values ​​may be generated in the image, i.e., transformation holes. To eliminate these holes and ensure image integrity, interpolation padding is required. For example, bilinear interpolation, bicubic interpolation, or more advanced image inpainting algorithms can be used to estimate and fill the gray values ​​of missing pixels based on the gray values ​​of surrounding known pixels, thereby obtaining multiple complete aligned images.

[0055] The proposed solution first constructs a precise pixel coordinate transformation mapping relationship based on a global translation vector and a global rotation angle, which quantifies and describes the overall motion trend of each subsequent frame relative to the reference frame. It is precisely this precise mapping relationship that enables subsequent pixel-level spatial transformations to accurately align the anatomical structure pixel positions of each subsequent frame with the corresponding anatomical structure pixel positions of the reference frame, thereby effectively eliminating inter-frame positional offset errors caused by vibration. Furthermore, interpolation filling is performed on any missing pixel positions that may appear after the spatial transformation, ensuring the integrity of the aligned image and preventing data loss from affecting subsequent image fusion operations.

[0056] The above technical solution enables precise motion-compensated alignment of multiple original X-ray images. Specifically, by constructing a pixel coordinate transformation mapping relationship and performing pixel-level spatial transformation, image translation and rotation caused by vibration of the portable DR detector can be effectively corrected, ensuring high consistency of the same anatomical structures at the pixel level between different frames. Furthermore, by interpolating and filling missing pixels after transformation, image information loss and visual artifacts are avoided, resulting in high-quality, hole-free, and fully aligned images. This provides a reliable foundation for subsequent pixel-level weighted fusion operations, significantly improving the clarity and diagnostic value of the final fused image.

[0057] In some embodiments described above in this application, pixel-level weighted fusion operations are performed on multiple aligned images to generate a fused image and suppress image motion blur caused by vibration information. Specifically, the aforementioned pixel-level weighted fusion operation on multiple aligned images may include the following steps: Calculate the local gray-level statistical features of each pixel position in each aligned image pixel by pixel, and quantize to obtain the noise level estimate of the corresponding pixel position; The fusion weights are adaptively assigned based on the noise level estimate of each pixel location, wherein the pixel location with the lower noise level estimate is assigned a higher fusion weight. The gray values ​​at the same pixel position in multiple aligned images are weighted and summed, and the weighted sum is used as the gray value of the corresponding pixel position in the fused image, thus completing pixel-level image fusion.

[0058] Specifically, when fusing multiple aligned images, it is first necessary to perform local analysis on each pixel location of each aligned image. Local gray-level statistical features can be understood as the distribution characteristics of gray-level values ​​within a small neighborhood around a pixel. For example, the gray-level variance, standard deviation, or mean absolute deviation within that neighborhood can be calculated. These statistical features can effectively reflect the image noise level in that local area. For instance, when the gray-level values ​​in a local area fluctuate significantly, it usually means that the area may be affected by strong noise or blurring, and the noise level estimate obtained by quantization will be relatively high.

[0059] The adaptive allocation of fusion weights based on the noise level estimates of each pixel location means dynamically adjusting the proportion of each pixel in the final fusion result based on its estimated noise level. Specifically, pixels with lower noise level estimates represent clearer image information with less noise, and therefore are given higher fusion weights, making them more dominant in the fused image. Conversely, pixels with higher noise level estimates are given lower fusion weights to mitigate their negative impact on the fusion result.

[0060] In practical applications, weighted summation of grayscale values ​​at the same pixel location in multiple aligned images involves linearly combining the grayscale values ​​at that pixel location from all aligned images according to their assigned fusion weights. For example, if there are N aligned images, and for a certain pixel location (x, y), its grayscale value in the i-th image is G_i(x, y), and the corresponding fusion weight is W_i(x, y), then the grayscale value G_f(x, y) of the fused image at that pixel location can be expressed as: G_f(x, y) = Σ [G_i(x, y) * W_i(x, y)] / Σ [W_i(x, y)]. In this way, information from multiple images can be effectively integrated to generate a fused image with a higher signal-to-noise ratio and less motion blur.

[0061] The proposed solution calculates the local gray-level statistical features of each aligned image pixel by pixel and quantizes them to obtain a noise level estimate, thereby enabling a precise assessment of image quality differences in different regions and frames. It is precisely this precise noise assessment that allows for adaptive adjustment of subsequent fusion weight allocation.

[0062] Through the above technical solution, this application can achieve refined fusion of multiple aligned images, rather than simple averaging or fixed-weight fusion. This adaptive weighted fusion based on pixel-level noise levels enables the fusion process to intelligently identify and utilize high-quality regions in each frame while effectively suppressing the negative impact of low-quality regions. Therefore, it can significantly improve the signal-to-noise ratio and sharpness of the fused image. Especially in portable DR imaging, where vibration causes differences in image quality across different frames or regions, this scheme can more effectively eliminate motion blur and noise, generating DR images with higher diagnostic value.

[0063] Specifically, in the aforementioned portable DR image intelligent enhancement method, the identified abnormal regions are judged to distinguish between real anatomical structures and image artifacts. The specific process includes: Extract the gradient information of the edge pixels in the abnormal region, and statistically obtain the gradient direction distribution characteristics of the edge pixels. If the gradient direction of the edge pixels is uniformly distributed in multiple directions, then the abnormal region is determined to be an image artifact. If the gradient direction of the edge pixels is concentrated in a single dominant direction, then the abnormal region is determined to be a real anatomical structure.

[0064] Extracting gradient information from edge pixels in anomaly regions involves using image processing algorithms, such as Sobel, Prewitt, or Roberts gradient operators, to calculate the rate of grayscale change of each pixel within the anomaly region in the horizontal and vertical directions, thereby obtaining the gradient magnitude and direction of that pixel. Gradient direction distribution characteristics can be understood as statistically analyzing the gradient directions of all edge pixels within the anomaly region to quantify their distribution in different directions. For example, this distribution can be characterized by constructing a gradient direction histogram or calculating the statistical variance and entropy of the gradient directions.

[0065] In practical applications, if the gradient directions of edge pixels are uniformly distributed in multiple directions, it usually means that the gray-level changes in that region lack regularity, and the edge directions are random and discontinuous. This is consistent with the characteristics of image artifacts caused by noise, motion blur, or X-ray scattering. For example, when there is random noise or slight motion blur in an image, the pixel gray-level changes in the resulting artifact regions are often disordered, resulting in a uniform distribution of edge gradient directions in all directions. Conversely, if the gradient directions of edge pixels are clustered in a single dominant direction, it indicates that the gray-level changes in that region have a clear directionality, and the edges are continuous and the structure is clear. This is usually an inherent characteristic of real anatomical structures (such as bone edges and blood vessel contours). For example, the edges of bones are usually smooth and continuous, and the gradient directions of their edge pixels will show a high degree of consistency along the contour direction of the bone.

[0066] This application's solution effectively distinguishes image artifacts from real anatomical structures by analyzing the gradient direction distribution characteristics of pixels at the edges of abnormal regions. Image artifacts, especially those caused by vibration or noise, typically exhibit random and irregular pixel grayscale variations, resulting in a disordered, multi-directional, uniformly distributed gradient direction at their edge pixels. In contrast, real anatomical structures, such as the edges of bones and organs, have clear and continuous boundaries. The gradient directions of pixels at these boundaries show consistency or specific clustering along the contour of the structure. Therefore, by quantifying and comparing these gradient direction distribution characteristics, irregular artifacts can be objectively distinguished from regular real anatomical structures.

[0067] Through the above technical solution, this application provides an objective discrimination method based on the intrinsic features of an image to accurately distinguish whether the abnormal regions identified in the fused image are real anatomical structures or image artifacts. This discrimination mechanism avoids the misjudgment that may occur if only local texture feature comparison is relied upon. Especially in DR images, various forms of artifacts may be introduced due to factors such as vibration. This method can effectively reduce false alarms and ensure that subsequent image optimization operations only target real artifacts, thereby protecting the integrity of real anatomical structures and improving the accuracy and reliability of image enhancement.

[0068] In some embodiments described above, the portable DR image intelligent enhancement method generates a fused image by performing pixel-level weighted fusion operations on multiple aligned images, and extracts local texture features from the fused image to identify abnormal regions. If the abnormal region is determined to be an image artifact, the weight coefficients corresponding to the pixel-level weighted fusion are adjusted, and the image fusion operation is re-executed to generate an optimized fused image. However, in practical applications, if the weight adjustment of artifact regions is not precise enough or lacks specificity, it may lead to incomplete artifact removal, or cause a certain degree of blurring or distortion of the real anatomical structure while removing artifacts, thereby affecting the diagnostic quality of the final image.

[0069] In response, this application further proposes specific steps to adjust the weight coefficients corresponding to pixel-level weighted fusion, re-execute the image fusion operation, and generate an optimized fused image. The aim is to eliminate image artifacts more effectively through a refined and localized weight adjustment strategy, while maximizing the preservation of the true anatomical details of the image.

[0070] Specifically, the above-mentioned adjustment of the weight coefficients corresponding to pixel-level weighted fusion, re-execution of the image fusion operation, and generation of an optimized fused image include: Locate abnormal regions in the fused image that belong to image artifacts, and extract the pixel positions of the corresponding regions to construct a set of artifact pixel positions; Reduce the fusion weight of the pixel positions corresponding to the artifact pixel position set in multiple aligned images; Pixel-level weighted fusion is re-executed based on the adjusted fusion weights to generate an optimized fused image with artifacts removed.

[0071] In this context, identifying abnormal regions that are image artifacts in the fused image refers to specific regions that have been clearly identified as image artifacts after the initial fused image generation, through texture feature comparison and artifact detection steps. Extracting the pixel locations of the corresponding regions to construct an artifact pixel location set specifically involves recording all pixel coordinates of these identified artifact regions in the image, forming a set. This set precisely identifies the image regions that require weight adjustment.

[0072] Furthermore, reducing the fusion weight of the pixel positions corresponding to the artifact pixel location set in multiple aligned images means reducing the weight of the corresponding pixel position in the multiple original aligned images for each pixel in the artifact pixel location set during pixel-level weighted fusion. For example, the magnitude of the weight reduction can be dynamically adjusted based on the severity of the artifact or its performance in different frame images. Typically, for original frame images that contribute significantly to the artifact region, the fusion weight of their corresponding pixel positions will be significantly reduced, or even set to zero, to suppress artifact generation to the greatest extent.

[0073] Therefore, re-performing pixel-level weighted fusion based on the adjusted fusion weights to generate an optimized fused image with artifacts removed means that after completing the aforementioned local weight adjustment, the system will perform pixel-level weighted fusion again. This fusion operation will use updated weight coefficients optimized for artifact regions to perform a weighted summation on multiple aligned images, thereby generating a new fused image. Because the weights of artifact regions are specifically reduced, the newly generated fused image will significantly reduce or eliminate the original image artifacts.

[0074] The proposed solution, after identifying image artifacts, does not simply re-fuse them. Instead, it first precisely locates the artifact regions and constructs a set of artifact pixel locations. Subsequently, for these artifact regions, the fusion weight of corresponding pixel locations is selectively reduced across multiple original aligned images. This localized and refined weight adjustment strategy ensures that when pixel-level weighted fusion is re-executed, the contribution of the original image data that caused the artifacts to the final fusion result is effectively suppressed. This avoids over-processing of non-artifact regions, thereby removing artifacts while preserving the detail and clarity of the true anatomical structures in the image to the maximum extent.

[0075] Through the above technical solution, this application can achieve precise localization and effective suppression of image artifacts, significantly improving the quality of portable DR images. This solution, by locally adjusting the weights of artifact regions, avoids indiscriminate re-fusion of the entire image, thus effectively protecting the true anatomical structural information of the image while removing artifacts. This results in an optimized fused image with higher diagnostic value and better visual effects. This targeted optimization strategy is particularly suitable for local artifacts in portable DR imaging that are susceptible to factors such as vibration, ensuring the accuracy and efficiency of image enhancement.

[0076] In some embodiments described above, this application proposes to regenerate an optimized fused image by adjusting the weight coefficients of pixel-level weighted fusion in a single step to suppress image artifacts. However, in practical applications, due to the complexity of vibration information or the subtlety of artifact characteristics during X-ray exposure imaging, a one-time weight adjustment may not completely eliminate all image artifacts, or residual, difficult-to-detect artifacts may still exist after adjustment, or even new artifacts may be introduced in some cases, resulting in room for improvement in the final enhanced image quality.

[0077] In this regard, this application further proposes that the above method also includes an iterative optimization step, which includes: The optimized fused image is regenerated, and local texture features of the new fused image are extracted. The re-extracted local texture features are compared a second time with the preset range of normal anatomical structure features; If no new abnormal regions are identified in the second comparison, the current fused image is output as the final enhanced image; If abnormal regions are still identified in the second comparison, the steps of adjusting the weight coefficients and regenerating the fused image are repeated until there are no abnormal regions.

[0078] Specifically, after adjusting the weight coefficients of pixel-level weighted fusion and generating an optimized fused image, the proposed solution does not end immediately but performs further quality evaluation on the optimized fused image. First, the optimized fused image needs to be regenerated, based on the result of the previous weight adjustment. Then, local texture features are extracted from the newly generated fused image, similar to the feature extraction process of the initial fused image, aiming to obtain the details and structural information contained in the image. Next, the re-extracted local texture features are compared a second time with a preset range of normal anatomical structure features. The purpose of this comparison is to re-check whether there are any abnormal regions in the image whose features deviate from the preset range, in order to determine whether there are residual or newly generated image artifacts.

[0079] If the second comparison shows no new anomalous regions, it means that the image quality has reached the expected level after the previous weight adjustment and fusion, and there are no longer any obvious artifacts or anomalous structures. In this case, the current fused image is output as the final enhanced image. Conversely, if the second comparison still identifies anomalous regions, it indicates that artifacts may still exist in the image, or the previous adjustment failed to completely solve the problem. In this case, the system will repeat the weight coefficient adjustment and fused image regeneration steps. This means that the weight coefficients of pixel-level weighted fusion will be adjusted again based on the newly identified anomalous regions, and the image fusion operation will be performed again to generate a new optimized fused image. This iterative process will continue until the second comparison no longer identifies any anomalous regions, thus ensuring that the final output enhanced image has higher quality and fewer artifacts.

[0080] The iterative optimization step is introduced in this application because in complex real-world imaging environments, a single artifact detection and correction may not fully address all issues. By introducing iterative optimization, this application enables multiple rounds of artifact detection, discrimination, and correction. Each iteration is based on the image optimized in the previous round, allowing the system to gradually approach the optimal image quality state. This mechanism allows the system to promptly readjust when residual or new artifacts are detected, thus avoiding image quality defects caused by insufficient processing in a single iteration. It is precisely this cyclical feedback and correction mechanism that enables this application to more comprehensively and thoroughly eliminate image artifacts and improve the overall image quality.

[0081] In some embodiments described above, pixel-level weighted fusion of multiple aligned images can generate a fused image and suppress motion blur caused by vibration. However, in practical applications, the vibration amplitude of portable DR detectors may fluctuate drastically during X-ray exposure imaging, causing some image frames to be acquired during peak vibration periods, resulting in severely degraded image quality. If only general pixel-level weighted fusion is relied upon, these severely degraded image frames may still interfere with the final fusion result, leading to residual artifacts or insufficient sharpness in the fused image. To address this, this application further proposes a fusion weight optimization step based on vibration periods, aiming to further improve the quality and robustness of the fused image by identifying and mitigating the impact of inferior image frames acquired during high vibration periods on the fusion result.

[0082] The above-mentioned fusion weight optimization steps based on vibration time periods include: When acquiring multiple raw X-ray images, the acquisition timestamps of each frame are recorded simultaneously. Determine the peak periods when the vibration amplitude exceeds a preset amplitude threshold based on vibration information; In the pixel-level weighted fusion process, the overall fusion weight of image frames whose acquisition timestamps fall within the peak period is reduced to weaken the interference of high-vibration, low-quality images on the fusion result.

[0083] Specifically, during the process of a portable DR detector performing X-ray exposure imaging and acquiring multiple raw X-ray images, the completion or start time of each image acquisition is precisely recorded, forming a corresponding acquisition timestamp. These timestamps are synchronized with the time series of vibration information for subsequent temporal correlation matching. The vibration information is typically acquired in real time by the detector's built-in accelerometer or other vibration monitoring devices. By analyzing this vibration information, specific time periods where the vibration amplitude is significantly higher than normal levels can be identified, i.e., peak periods.

[0084] For example, a preset amplitude threshold can be set. When the vibration amplitude continuously or intermittently exceeds this threshold, the corresponding time period is marked as the peak period. This threshold can be set according to the actual application scenario and experience to distinguish between normal vibration and severe vibration that may cause serious image degradation. In practical applications, when performing pixel-level weighted fusion operations on multiple aligned images, the acquisition timestamp of each frame is first checked to see if it falls within the determined peak period. For image frames whose acquisition timestamps are within the peak period, their overall fusion weight in the fusion operation will be significantly reduced. This reduction can be a uniform scaling factor or an adaptive adjustment based on the deviation of the vibration amplitude from the threshold. The purpose is to weaken the contribution of these inferior image frames acquired under severe vibration to the final fused image from a global perspective, thereby effectively suppressing image artifacts and blurring caused by high vibration and ensuring the overall quality of the fused image.

[0085] This application's solution refines the traditional pixel-level weighted fusion mechanism by introducing the concept of vibration periods. While the basic solution suppresses motion blur through pixel-level weighted fusion, its weight allocation is primarily based on the grayscale statistical characteristics of local pixels, which may not be sufficient to identify and suppress image frames that are severely degraded overall due to intense vibration. This solution, however, precisely associates image frames with specific vibration periods by synchronously recording image acquisition timestamps and vibration information. When a peak period with vibration amplitude exceeding a preset threshold is identified, image frames acquired during that period are determined to have a high risk of degradation. Therefore, by reducing the overall fusion weight of these high-risk image frames, their negative impact on the final fused image can be reduced from the source, even after motion compensation, as residual artifacts or noise may still exist that are difficult to completely eliminate. This time-dimensional weight adjustment mechanism allows the fusion process to more intelligently avoid the worst-quality data, thereby improving the robustness and clarity of the fusion result.

[0086] In some of the embodiments described above in this application, techniques such as multi-frame exposure, motion compensation alignment, and pixel-level weighted fusion can effectively suppress image motion blur caused by vibration and identify and correct image artifacts. However, during X-ray exposure imaging with a portable DR detector, if the vibration amplitude is too large, the acquisition quality of the original X-ray image may be inherently low, such as a decreased signal-to-noise ratio or insufficient contrast. This increases the difficulty of subsequent image processing and may even affect the diagnostic effect of the final enhanced image. If the image acquisition quality is not optimized from the source, even the most refined subsequent processing may not be able to completely compensate for the inherent defects of the original image.

[0087] In response, this application further proposes an adaptive optimization step for exposure parameters, which includes: Obtain the current exposure dose parameters of the portable DR detector; Based on the vibration amplitude value of the vibration information, the exposure dose adjustment factor is determined according to the preset mapping relationship between amplitude and dose, wherein the vibration amplitude value and the exposure dose adjustment factor are monotonically increasing in the mapping relationship; Multiply the current exposure dose parameter by the exposure dose adjustment factor to obtain the optimized exposure dose parameter, which is used to optimize the quality of the original image acquisition.

[0088] Specifically, acquiring the current exposure dose parameter of a portable DR detector refers to the system automatically reading or receiving the X-ray exposure dose value set by the DR detector before or during each X-ray exposure imaging. This dose parameter is usually expressed in units such as milliampere-seconds (mAs) or milliroentgens (mR), and directly affects the number of photons that reach the detector after X-rays penetrate the object.

[0089] The vibration amplitude value of the vibration information can be understood as a quantitative indicator of the vibration intensity of the portable DR detector during exposure, acquired in real-time or near real-time by the vibration monitoring module, such as peak acceleration and root mean square acceleration. The preset amplitude-dose mapping relationship refers to a set of rules or lookup tables pre-stored or configured within the system, which defines the adjustment factor of the exposure dose corresponding to different vibration amplitude values.

[0090] In a preferred embodiment, the vibration amplitude value and the exposure dose adjustment factor in the mapping relationship are monotonically increasing. This means that the greater the vibration amplitude, the greater the required exposure dose adjustment factor, in order to compensate for the potential decrease in image quality due to vibration. Multiplying the current exposure dose parameter by the exposure dose adjustment factor yields the optimized exposure dose parameter. This involves multiplying the originally set exposure dose by the adjustment factor determined according to the mapping relationship to obtain a new exposure dose more suitable for the current vibration conditions. This optimized exposure dose parameter is then sent to the X-ray generator to guide the actual X-ray exposure process, aiming to optimize the quality of the original image acquisition and ensure high-quality original X-ray images are obtained under different vibration environments.

[0091] The proposed solution dynamically adjusts the exposure dose parameter during X-ray exposure imaging based on the vibration amplitude value acquired in real time by a portable DR detector, thereby actively optimizing the acquisition quality of the original image. When the vibration amplitude increases, the system correspondingly increases the exposure dose, which helps increase the number of X-ray photons reaching the detector, thus improving the signal-to-noise ratio of the original image and reducing signal loss and noise interference that may be caused by vibration. Therefore, even in environments with severe vibration, the system ensures that multiple acquired original X-ray images have good basic quality, providing higher-quality input data for subsequent processing steps such as motion compensation, image fusion, and artifact detection, effectively overcoming the limitations that may exist if post-processing is relied upon alone.

[0092] In some preferred embodiments, it is assumed that the portable DR detector's vibration monitoring module detects a current vibration amplitude of 0.8g during X-ray exposure. The system's internal preset amplitude-dose mapping specifies that when the vibration amplitude is 0.8g, the exposure dose adjustment factor is 1.15. If the current exposure dose parameter is set to 5 mAs, the system will multiply 5 mAs by 1.15 to obtain an optimized exposure dose parameter of 5.75 mAs. Subsequently, the X-ray generator will expose the object at a dose of 5.75 mAs, thereby acquiring a raw X-ray image with a higher signal-to-noise ratio and better quality under the current vibration conditions.

[0093] refer to Figure 3 , Figure 3 This is a schematic diagram of the structure of a portable DR image intelligent enhancement system provided in an embodiment of the present invention. The system includes: a vibration monitoring module, an exposure control module, a motion processing module, an image fusion module, an artifact detection module, an artifact discrimination module, and a fusion adjustment module. The vibration monitoring module is used to acquire vibration information generated by the portable DR detector during X-ray exposure imaging. The exposure control module is used to adjust the X-ray exposure mode to a multi-frame exposure mode based on the vibration information and to acquire multiple raw X-ray images. The motion processing module is used to calculate the inter-frame relative motion vector based on the pixel grayscale matching results of multiple original X-ray images, and to perform motion compensation and alignment processing on the original images based on the relative motion vector to obtain multiple aligned images. The image fusion module is used to perform pixel-level weighted fusion of multiple aligned images to generate a fused image and suppress motion blur caused by vibration. The artifact detection module is used to extract local texture features from the fused image, compare the local texture features with a preset range of normal anatomical structure features, and identify abnormal areas in the image. The artifact discrimination module is used to determine whether an abnormal region is a real anatomical structure or an image artifact based on the edge gradient distribution characteristics of the abnormal region. The fusion adjustment module is used to adjust the weight coefficients of pixel-level weighted fusion when image artifacts are determined to exist, and regenerate the optimized fused image to achieve image enhancement optimization.

[0094] The portable DR image intelligent enhancement system proposed in this application aims to solve the problems of motion blur caused by vibration in traditional portable DR devices during imaging in complex environments, and the potential for clinically misleading artifacts in existing enhancement technologies. This system integrates multiple functional modules, including vibration monitoring, exposure control, motion processing, image fusion, artifact detection, artifact discrimination, and fusion adjustment, forming a closed-loop intelligent processing flow. Specifically, the system first senses vibration during the imaging process in real time and dynamically adjusts the exposure strategy accordingly to acquire multiple frames of images. Subsequently, it performs precise motion compensation and fusion on these images to effectively suppress motion blur. More importantly, the system can intelligently identify and distinguish whether abnormal areas in the fused image are real anatomical structures or image artifacts, and adaptively adjust the fusion parameters based on the discrimination results to regenerate an optimized image. This fundamentally avoids the risk of misdiagnosis and significantly improves the diagnostic quality and safety of portable DR images.

[0095] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A portable intelligent image enhancement method for DR (Digital Radiography), characterized in that, The method includes: Acquire vibration information generated by a portable DR detector during X-ray exposure imaging; Based on the vibration information, the X-ray exposure mode is adjusted to a multi-frame exposure mode, and multiple raw X-ray images are acquired. Based on the pixel grayscale matching results between the multiple original X-ray images, the relative motion vector between each frame image is calculated. Based on the relative motion vector, motion compensation and alignment processing is performed on the multiple original X-ray images to eliminate inter-frame position offset errors and obtain multiple aligned images. Pixel-level weighted fusion operations are performed on the multiple aligned images to generate a fused image, thereby suppressing image motion blur caused by vibration information; Local texture features of the fused image are extracted and compared with a preset range of normal anatomical structure features. Abnormal regions where the features deviate from the preset range are identified, and the abnormal regions are determined to be real anatomical structures or image artifacts. Gradient information of edge pixels of the abnormal regions is extracted, and the gradient direction distribution features of the edge pixels are statistically obtained. If the gradient direction of the edge pixels is uniformly distributed in multiple directions, the abnormal region is determined to be an image artifact; if the gradient direction of the edge pixels is clustered in a single dominant direction, the abnormal region is determined to be a real anatomical structure. If the abnormal region is determined to be an image artifact, the weight coefficients corresponding to the pixel-level weighted fusion are adjusted, the image fusion operation is re-executed, and an optimized fused image is generated.

2. The portable DR image intelligent enhancement method according to claim 1, characterized in that, The calculation of the relative motion vectors between each frame image includes: One frame is selected from the multiple original X-ray images as a reference frame, and multiple pixels with gray-scale gradient changes are extracted from the reference frame as feature tracking points. In the remaining subsequent frames, pixel coordinates of each feature tracking point are tracked one by one to obtain the displacement vector corresponding to each tracking point. By integrating the displacement vector data of all feature tracking points, the global translation vector and global rotation angle between the reference frame and each subsequent frame are obtained through vector fitting, thus obtaining the inter-frame relative motion vector.

3. The portable DR image intelligent enhancement method according to claim 2, characterized in that, The motion-compensated alignment process for the multiple original X-ray images includes: Based on the global translation vector and the global rotation angle, construct the pixel coordinate transformation mapping relationship between each subsequent frame image and the reference frame; According to the coordinate transformation mapping relationship, perform pixel-level spatial transformation on each subsequent frame image so that the anatomical structure pixel position of each subsequent frame is aligned with the corresponding anatomical structure pixel position of the reference frame. Interpolation is performed on the missing pixel positions in the spatially transformed image to eliminate transformation holes and obtain multiple complete aligned images.

4. The portable DR image intelligent enhancement method according to claim 1, characterized in that, The pixel-level weighted fusion operation on the multiple aligned images specifically includes: Calculate the local gray-level statistical features of each pixel position in each aligned image pixel by pixel, and quantize to obtain the noise level estimate of the corresponding pixel position; The fusion weights are adaptively assigned based on the noise level estimate of each pixel location, wherein the pixel location with the lower noise level estimate is assigned a higher fusion weight. The gray values ​​at the same pixel position in multiple aligned images are weighted and summed, and the weighted sum is used as the gray value of the corresponding pixel position in the fused image, thus completing pixel-level image fusion.

5. The portable DR image intelligent enhancement method according to claim 1, characterized in that, The step of adjusting the weight coefficients corresponding to pixel-level weighted fusion, re-executing the image fusion operation, and generating an optimized fused image includes: Locate abnormal regions in the fused image that belong to image artifacts, and extract the pixel positions of the corresponding regions to construct a set of artifact pixel positions; Reduce the fusion weight of the pixel positions corresponding to the artifact pixel position set in multiple aligned images; Pixel-level weighted fusion is re-executed based on the adjusted fusion weights to generate an optimized fused image with artifacts removed.

6. The portable DR image intelligent enhancement method according to claim 1, characterized in that, The method further includes an iterative optimization step, which includes: The optimized fused image is regenerated, and local texture features of the new fused image are extracted. The re-extracted local texture features are compared a second time with the preset range of normal anatomical structure features; If no new abnormal regions are identified in the second comparison, the current fused image is output as the final enhanced image; If abnormal regions are still identified in the second comparison, the steps of adjusting the weight coefficients and regenerating the fused image are repeated until there are no abnormal regions.

7. The portable DR image intelligent enhancement method according to claim 1, characterized in that, The method further includes a fusion weight optimization step based on vibration time periods, which includes: When acquiring the multiple original X-ray images, the acquisition timestamp of each frame is recorded synchronously. The peak period during which the vibration amplitude exceeds a preset amplitude threshold is determined based on the vibration information; In the pixel-level weighted fusion process, the overall fusion weight of image frames whose acquisition timestamps fall within the peak period is reduced to weaken the interference of high-vibration, low-quality images on the fusion result.

8. The portable DR image intelligent enhancement method according to claim 1, characterized in that, The method further includes an exposure parameter adaptive optimization step, which includes: Obtain the current exposure dose parameters of the portable DR detector; Based on the vibration amplitude value of the vibration information, the exposure dose adjustment factor is determined according to the preset mapping relationship between amplitude and dose, wherein the vibration amplitude value and the exposure dose adjustment factor are monotonically increasing in the mapping relationship; The current exposure dose parameter is multiplied by the exposure dose adjustment factor to obtain the optimized exposure dose parameter, which is used to optimize the quality of the original image acquisition.

9. A portable DR image intelligent enhancement system, characterized in that, The system is used to perform the portable DR image intelligent enhancement method according to any one of claims 1 to 8, the system comprising: a vibration monitoring module, an exposure control module, a motion processing module, an image fusion module, an artifact detection module, an artifact discrimination module, and a fusion adjustment module; The vibration monitoring module is used to acquire vibration information generated by the portable DR detector during X-ray exposure imaging. The exposure control module is used to adjust the X-ray exposure mode to a multi-frame exposure mode based on the vibration information and to acquire multiple raw X-ray images. The motion processing module is used to calculate the inter-frame relative motion vector based on the pixel grayscale matching results of multiple original X-ray images, and to perform motion compensation and alignment processing on the original images based on the relative motion vector to obtain multiple aligned images. The image fusion module is used to perform pixel-level weighted fusion of multiple aligned images to generate a fused image and suppress motion blur caused by vibration. The artifact detection module is used to extract local texture features from the fused image, compare the local texture features with a preset range of normal anatomical structure features, and identify abnormal areas in the image. The artifact discrimination module is used to determine whether an abnormal region is a real anatomical structure or an image artifact based on the edge gradient distribution characteristics of the abnormal region. It extracts the gradient information of the edge pixels of the abnormal region and statistically obtains the gradient direction distribution characteristics of the edge pixels. If the gradient direction of the edge pixels is uniformly distributed in multiple directions, the abnormal region is determined to be an image artifact. If the gradient direction of the edge pixels is clustered in a single dominant direction, the abnormal region is determined to be a real anatomical structure. The fusion adjustment module is used to adjust the weight coefficients of pixel-level weighted fusion when image artifacts are determined to exist, and regenerate the optimized fused image to achieve image enhancement optimization.

Images