Description:TITLE - AI-POWERED X-RAY IMAGE ENHANCEMENT

TECHNICAL FIELD
The present disclosure relates generally to the field of digital image processing. More particularly, the present disclosure pertains to systems and methods for enhancing medical images. Specifically, the present disclosure relates to using deep learning models for real-time magnification, denoising, and enhancement of radiological images, such as those generated during X-ray fluoroscopy.

BACKGROUND
X-ray fluoroscopy is a critical medical imaging technique that provides real-time visualization of internal anatomical structures, facilitating complex procedures such as angiography, catheter placements, and orthopedic evaluations. During such procedures, clinicians often need to magnify a region of interest (ROI) to visualize fine anatomical details for improved diagnostic accuracy and procedural guidance. However, achieving high-quality magnification in real-time presents significant challenges.
Existing methods for image magnification in X-ray systems typically fall into two categories. One approach involves hardware-based magnification, which enhances image quality by increasing the X-ray dosage. While this can produce clearer images, it disadvantageously exposes both the patient and the clinical staff to additional, often significant, amounts of radiation. The other common approach relies on software-based interpolation methods, such as Lanczos or bicubic interpolation, to enlarge the image digitally.
These classical software interpolation methods, however, struggle to adequately handle the noisy and random nature inherent in X-ray images. This often results in suboptimal image quality, where the magnified image may appear blurry, lack fine details, or suffer from artifacts, thereby limiting its clinical utility. Consequently, there remains a need in the field for a system and method that can provide real-time, high-quality magnification and enhancement of X-ray images while simultaneously minimizing radiation exposure and effectively overcoming the limitations of traditional software-based techniques.

SUMMARY
The present disclosure relates generally to digital image processing, and more particularly, the present disclosure relates to a method, system, and non-transitory computer-readable medium for real-time enhancement of X-ray images using a unified deep learning model.
It is an object of the present disclosure to provide an improved method and system for real-time X-ray image enhancement. Moreover, the present disclosure relates to a method and system that uses a unified deep learning model to simultaneously perform super-resolution, blind denoising, and image enhancement on a low-resolution X-ray image in a single processing step. Further, the present disclosure relates to a non-transitory computer-readable medium that includes instructions for carrying out the method, when said instructions are executed on a computer system.
This object is achieved by the features of the various aspects of the disclosure described herein. Further, implementation forms are apparent from the description and the figures.
According to a first aspect, there is provided a method for real-time enhancement of an X-ray image. The method includes receiving a low-resolution X-ray image comprising noise. The method includes processing, via one or more processors, the low-resolution X-ray image with a unified deep learning model configured to, in a single processing step, simultaneously perform super-resolution, blind-denoising, and image enhancement to thereby generate a high-resolution, enhanced image. The model comprises a sequence of integrated layers configured to first, extract a feature map via a feature extraction layer; second, refine the feature map via one or more intermediate layers including an attention layer configured to apply an attention map; and third, reconstruct the high-resolution, enhanced image from the refined and attention-modulated feature map via an upsampling layer. The method concludes with providing the high-resolution, enhanced image for display.
Preferably, the attention layer computes the attention map using a convolution operation followed by a sigmoid activation function. More preferably, the attention-modulated feature map is generated by performing an element-wise multiplication of a refined feature map with the attention map.
Preferably, the one or more intermediate layers further comprise a shrinking layer, a mapping layer, and an expanding layer, arranged in sequence.
Preferably, the upsampling layer reconstructs the high-resolution, enhanced image using sub-pixel convolution.
Preferably, the unified deep learning model is trained by minimizing a composite loss function that balances pixel-wise accuracy, noise suppression, and perceptual quality. More preferably, this composite loss function comprises a pixel loss component, a perceptual loss component, and a pyramid loss component.
According to a preferred embodiment, the unified deep learning model executed by the one or more processors is integrated within an x-ray imaging device, such that the enhancement of the low-resolution x-ray image is performed onboard said x-ray imaging device prior to providing the enhanced image for display.
According to a second aspect, there is provided a system comprising one or more processors and a memory storing instructions for a unified deep learning model that, when executed by the one or more processors, configure the system to carry out all the steps of the above-described method.
Preferably, the one or more processors and the memory are integrated within an x-ray imaging device such that the unified deep learning model executes locally within said x-ray imaging device to provide real-time enhancement of image data acquired by the x-ray imaging device.
According to a third aspect, there is provided a non-transitory computer-readable medium storing instructions thereon that, when executed by one or more processors, cause a system to perform the above-described method.
The method, system, and computer-readable medium described herein provide several benefits due to their design and technical principles, overcoming limitations in existing X-ray imaging techniques.
The described approach offers superior image quality for magnified regions of interest in real-time. The unified, single-step model architecture allows for extremely efficient processing, making it suitable for live clinical workflows like X-ray fluoroscopy. By simultaneously performing super-resolution and blind denoising, the method provides high-quality magnification without requiring an increase in X-ray dosage, thereby significantly reducing radiation exposure for both patients and clinical staff.
A key advantage is the improvement in diagnostic clarity. The use of attention layers enables the model to focus on clinically relevant features, while the composite loss function, particularly the pyramid loss component, ensures the enhancement is robust across multiple zoom levels. This enhanced clarity can reduce the need for additional contrast dye injections, improving patient safety by lessening the renal burden.
Furthermore, the modular architecture, which is trained using a sophisticated loss function, results in a robust system capable of handling the inherent noise and randomness of X-ray images, a common failure point for traditional interpolation-based methods.
Therefore, in contradistinction to existing solutions that either increase radiation exposure or produce suboptimal, blurry images, the described method, system, and computer-readable medium provide a safer and more effective solution for real-time X-ray image enhancement, leading to improved diagnostic accuracy and enhanced patient outcomes.
These and other aspects of the disclosure will be apparent from the implementation(s) described below.

BRIEF DESCRIPTION OF DRAWINGS
Implementations of the disclosure will now be described, by way of example only, with reference to the accompanying drawings, in which:
FIG. 1 is a block diagram illustrating an exemplary system for real-time enhancement of an X-ray image, in accordance with an implementation of the present disclosure.
FIG. 2 is a block diagram illustrating the internal architecture of the Unified Deep Learning Model, featuring the specific sequence of layers, in accordance with an implementation of the present disclosure.
FIG. 3 is a flowchart illustrating a method for real-time enhancement of an X-ray image, in accordance with an implementation of the present disclosure.
FIG. 4 illustrates a comparison of images showing an exemplary result of the X-ray image enhancement process, in accordance with an implementation of the present disclosure.
FIG. 5 is a block diagram illustrating the computer system architecture for implementing the system for real-time enhancement of an X-ray image.

DETAILED DESCRIPTION
Implementations of the present disclosure provide a system and method for real-time X-ray image enhancement using a unified deep learning model for simultaneous super-resolution, blind denoising, and image enhancement, implemented within a medical imaging data processing system. This enables improved diagnostic clarity, reduced radiation exposure for patients and staff, and application to tasks such as angiography, catheter placements, and orthopedic evaluations. Moreover, the present disclosure relates to a system for performing X-ray image processing through the enhancement of a low-resolution image via a single-pass deep learning model that reconstructs high-frequency details while suppressing noise. Further, the present disclosure relates to a computer program that includes instructions for carrying out the image enhancement method, when said computer program is executed on a computer system.
The disclosed method and system address limitations of existing techniques by enabling high-quality, real-time image magnification through a robust and efficient deep learning model. This approach offers superior image quality in challenging real-time clinical workflows, demonstrates robustness against inherent image noise and randomness, and provides the potential for elimination of additional radiation exposure and contrast dye usage. The disclosed techniques facilitate a unified, single-step processing architecture, support configurable magnification and flexible display positioning, and offer seamless integration with existing X-ray systems, with performance often evaluated using metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Metric (SSIM). To make implementations of the present disclosure more comprehensible for a person skilled in the art, the following implementations are described with reference to the accompanying drawings, including FIG. 1 which illustrates an exemplary system block diagram for real-time X-ray image enhancement, FIG. 2 which depicts a detailed block diagram of the internal architecture of the unified deep learning model, and FIG. 3 which shows a flowchart of an exemplary method for real-time X-ray image enhancement.
Terms such as "a first", "a second", "a third", and "a fourth" (if any) in the summary, claims, and foregoing accompanying drawings of the present disclosure are used to distinguish between similar objects and are not necessarily used to describe a specific sequence or order. It should be understood that the terms so used are interchangeable under appropriate circumstances, so that the implementations of the present disclosure described herein are, for example, capable of being implemented in sequences other than the sequences illustrated or described herein. Furthermore, the terms "include" and "have" and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, a method, a system, a product, or a device that includes a series of steps or units for image processing, is not necessarily limited to expressly listed steps or units but may include other steps or units that are not expressly listed or that are inherent to such process, method, product, or device.
The present disclosure provides a system and method for real-time enhancement of medical images, designed to overcome the challenges of traditional X-ray magnification techniques. The disclosure addresses the dual problems of increased radiation exposure from hardware-based approaches and the suboptimal image quality from classical software interpolation methods, which struggle with the inherent noise and stochastic nature of X-ray images. The described system provides a unified deep learning model for simultaneous super-resolution, blind denoising, and image enhancement in a single, real-time processing step.
FIG. 1 illustrates a block diagram of an exemplary system 100 for real-time enhancement of an X-ray image. The system 100 operates in conjunction with external peripherals, including an X-ray imaging device 102 and a display 104. The X-ray imaging device 102 is a medical imaging source, such as an X-ray fluoroscopy machine, configured to capture and provide a stream of low-resolution, noisy X-ray images as input. The display 104 is a clinical-grade monitor configured to receive and present the final, enhanced image to a medical professional for real-time visualization and diagnosis. The system 100 comprises internal hardware components, including one or more processors 106 (hereinafter, referred to as the “processor 106”) and a memory 108. The processor 106, which may include a central processing unit (CPU) and a graphics processing unit (GPU), serves as the computational engine of the system 100, responsible for executing the instructions stored in memory 108. The memory 108 is a computer-readable medium, such as RAM, that stores the executable software logic and temporarily holds image data during processing.
Residing within memory 108 is an enhancement module 110, which is the primary software component containing the complete logic for the X-ray image enhancement. The enhancement module 110 comprises several functional sub-modules: an image acquisition interface 112, a unified deep learning model 114, an optional post-processing module 116, and a display interface 118. In operation, the image acquisition interface 112 is configured to receive the low-resolution image data from the X-ray imaging device 102. The data is then processed by the unified deep learning model 114, which is the core neural network executed by the processor 106 to simultaneously perform super-resolution, blind denoising, and image enhancement in a single pass. After this core processing, the image may be optionally refined by the post-processing module 116, which applies further filters such as contrast adjustment, sharpening, or artifact removal. Finally, the display interface 118 formats the resulting high-resolution image and transmits it to the display 104 for visualization.
The system 100 is configured with a plurality of features to ensure robust clinical performance and usability. The system 100 is designed for real-time processing, capable of processing each image frame in less than 10 milliseconds on standard hardware configurations. It provides multi-resolution support, adeptly handling various input and output resolutions which can be set via user configuration. The system also allows for configurable magnification, enabling a clinical operator to specify zoom levels for precise visualization of a region of interest (ROI) through a user interface. For enhanced usability in a clinical setting, the system 100 further provides for flexible output window positioning, allowing the operator to move and place the window displaying the enhanced image anywhere on the screen. To support archival and subsequent review, the system features DICOM integration, which enables the saving of enhanced images or image sequences as DICOM loop files. The system 100 may be provided under flexible licensing models, such as a standalone license or as a fully integrated module within existing imaging software.
These system 100 features deliver significant clinical benefits. The primary benefit is a marked reduction in radiation exposure, as the disclosure's method of high-quality magnification negates the need for additional, high-dose X-ray acquisitions. This directly improves safety for both the patient and medical staff. The system 100 provides tangibly improved image quality, delivering sharper and clearer images than traditional interpolation methods, which is critical for accurate diagnosis. The real-time performance enhances workflow efficiency by enabling immediate decision-making during complex procedures. Furthermore, by enhancing image clarity to a degree that can obviate the need for additional contrast agents, the system 100 improves patient safety by reducing the renal burden associated with such agents. This leads to cost efficiency by extending the lifespan of x-ray equipment, such as the generator tubes, and reducing maintenance costs. The enhanced visualization capabilities also aid in the advanced diagnosis of difficult clinical conditions, including but not limited to Air Embolism, Ostial Stenosis, vascular Trifurcations/Bifurcations, and Dissections or Perforations.
FIG. 2 illustrates a block diagram of the internal architecture 200 of the unified deep learning model 114, featuring a specific sequence of layers that process the image data to achieve the desired enhancement. This architecture is the core of the disclosure, enabling the simultaneous performance of super-resolution, blind denoising, and image enhancement in a single, cohesive pass. The process within the model begins with the input of a low-resolution image (ILR) 202, which is the image data received by the image acquisition interface 112.
The input image 202 is first processed by a feature extraction layer 204. This initial layer is configured to extract key spatial features from the input image. It employs a convolution operation followed by a rectified linear unit (ReLU) activation function, a process mathematically described by the formula:
F="ReLU" (W_f*I_LR+b_f )
In this formulation, ILR represents the low-resolution input image, Wf is the convolution kernel for this layer, bf is an associated bias term, and F is the resulting output, referred to as the feature map. This feature map (F) carries the foundational details extracted from the original image. As used herein, the "bias term" refers to an adjustable scalar value added to the result of a convolution operation within a neural network layer, providing the model with an additional degree of freedom to fine-tune the output of that layer and thereby enabling more accurate feature detection and image enhancement during processing.
The feature map (F) is then passed to a shrinking layer 206. The function of the shrinking layer 206 is to reduce the dimensionality, or depth, of the feature map, thereby focusing the model's computation on the most essential features and improving efficiency. This is achieved using a 1x1 convolution followed by a ReLU activation function, formulated as:
S="ReLU" (W_s*F+b_s ),
where Ws and bs are the respective convolution kernel and bias term for the shrinking layer, and S is the resulting shrunk feature map.
Following the dimensionality reduction, the shrunk feature map (S) is fed into a mapping layer 208. The mapping layer 208 is responsible for further refining the features through one or more convolution operations, which prepares the features for the critical attention mechanism that follows. The operation is described by the formula:
M="ReLU" (W_m*S+b_m ),
where Wm and bm are the kernel and bias for the mapping layer, and M is the output, a refined feature map.
The refined feature map (M) is then processed by an attention layer 210. The attention layer 210 enables the model to intelligently focus on the most diagnostically relevant regions of the image while suppressing irrelevant information or noise. This is a two-step process. First, an attention map, A(x), is computed using a convolution operation followed by a sigmoid activation function, as shown by the formula:
A(x)=σ(W_a*M+b_a ),
where Wa is the convolutional kernel for the attention layer, ba is the bias term and σ is the sigmoid activation function.
Second, the attention-modulated feature map (Matt) is generated by performing an element-wise multiplication of the refined feature map with the attention map, a process described by the formula:
M_"att" =A(x)⋅M
Next, an expanding layer 212 takes the attention-modulated feature map (Matt) as input. The purpose of this layer is to restore the feature map's depth in preparation for the upsampling process, effectively reversing the action of the shrinking layer. This is accomplished using a 1x1 convolution followed by a ReLU activation, formulated as:
E="ReLU" (W_e*M_"att" +b_e ),
where We is the convolution kernel for the expanding layer, be is the bias term and E is the resulting expanded feature map.
The expanded feature map (E) is then passed to an upsampling layer 214. The upsampling layer 214 performs the super-resolution task by increasing the spatial resolution of the feature map by a specified scaling factor, r. The operation is represented as:
I_HR="Upsample" (E,"scale factor" =r),
where IHR is the resulting high-resolution image. This image may then be passed through a final convolution layer 216. This layer applies a final convolution, described by:
I_HR=W_c*I_HR+b_c,
to produce the final high-resolution image 218, ensuring it has the appropriate depth and refined details for output. In this formulation, Wc represents the kernel for the final convolution layer and bc represents the bias term. This entire sequence of layers, noted as being types of convolutional layers, allows the model to holistically transform the input image.
It is to be explicitly noted that the architecture depicted in FIG. 2 and its associated processing sequence are not merely an abstract mathematical method or a computer program "per se" but form a critical part of a technical system integrated with real-world hardware for achieving a tangible technical effect. Specifically, the unified deep learning model 114 is configured for deployment on specialized hardware platforms comprising processors such as GPUs and memory resources, and is tightly integrated into a medical imaging workflow involving real-time interaction with an X-ray imaging device and a clinical-grade display. The arrangement of feature extraction, shrinking, mapping, attention, expanding, and upsampling layers operates synergistically to process streaming medical image data in real-time — achieving end-to-end latency of less than 10 milliseconds — which is essential for safe and effective fluoroscopic procedures. This architecture produces a concrete technical improvement by delivering enhanced image resolution and diagnostic clarity without increasing X-ray dose levels, thus directly reducing radiation exposure to patients and clinical staff. Moreover, this improvement enables physicians to obtain clinically usable magnified images that would otherwise require repeat X-ray acquisitions or contrast dye injections, resulting in further technical and patient safety benefits. Accordingly, the unified deep learning model 114 as described and illustrated is part of a larger image processing apparatus that is integrated with external hardware devices, configured for a specific technical purpose, and achieves a real-world technical effect, thereby clearly distinguishing it from a software algorithm or abstract computer program.
FIG. 3 is a flowchart illustrating the overall method for real-time enhancement of an X-ray image, in accordance with an implementation of the present disclosure. The method begins at step 300, where a low-resolution X-ray image (ILR) comprising noise is received. This step is performed by the image acquisition interface 112 of the system 100, which is configured to intercept the real-time image stream from the X-ray imaging device 102. The data collection and preprocessing methodology are foundational to the model's performance. The model is trained on a comprehensive, proprietary dataset of high-resolution x-ray images paired with their low-resolution, noisy counterparts. This dataset is rigorously pre-processed, including normalization to standardize image intensities, data augmentation via rotations and flips to enhance model generalization, and the simulation of low-resolution, noisy versions from high-resolution source images to create training pairs. The model's training is guided by minimizing a composite loss function that balances three key aspects of image quality: a pixel loss for pixel-wise accuracy, a perceptual loss to ensure visual fidelity by comparing high-level features, and a pyramid loss to evaluate quality across multiple image scales. To optimize this training process, loss function weights are empirically determined via grid search and refined using validation metrics such as SSIM and PSNR, while strategies such as regularization, early stopping, and learning rate decay are employed to avoid overfitting and ensure generalization.
At step 302, the received image is processed by the unified deep learning model 114. As detailed in the description of FIG. 2, this model executes a single, unified inference pass to simultaneously perform super-resolution (302A), blind denoising (302B), and image enhancement (302C). This core processing step is fully automated and leverages the model's sophisticated architecture to reconstruct fine details and suppress noise, resulting in the generation of an intermediate high-resolution, enhanced image at step 304.
Following the core processing, the method proceeds to a decision step 306, where a determination is made whether to perform optional post-processing. This step provides clinical flexibility, allowing a system administrator to enable or disable specific post-processing operations via a separate settings file. If the decision is affirmative (the 'YES' path), the method moves to step 308, where the post-processing module 116 applies further filters to the intermediate image. These filters can include contrast adjustment to improve the visibility of anatomical structures, sharpening to enhance edge definition for clearer regions of interest, and artifact removal to suppress any residual noise or distortions.
Finally, at step 310, the final high-resolution, enhanced image (IHR)—either directly from step 304 if post-processing is bypassed (the 'NO' path), or after refinement from step 308—is provided for display. This step is handled by the display interface 118, which formats the image and sends it to the display 104 for real-time visualization. This step may also include saving the enhanced image sequence as a DICOM loop file for archival and later review, leveraging the system's DICOM integration feature.
FIG. 4 illustrates a visual comparison of images, showing an exemplary result of the real-time enhancement method applied to an x-ray angiography image, in accordance with an implementation of the present disclosure. FIG. 4 comprises four panels that demonstrate the progressive enhancement and magnification capabilities of the system 100. The leftmost panel may represent an original low-resolution input image or a simple digital zoom of a region of interest, characterized by a comparative lack of sharpness and visible noise. The subsequent panels, moving from left to right, show the output of the enhancement module 110 at increasing levels of magnification and enhancement. A clear visual improvement can be observed across the panels, demonstrating the effectiveness of the unified deep learning model 114. The vessel edges appear significantly sharper, the contrast between the vessels and the surrounding anatomy is improved, and the inherent image noise is suppressed. This allows for finer details in the smaller arterial branches to become visible, which is critical for accurate diagnosis and supports the clinical benefits of the disclosure.
FIG. 5 is an illustration of a computer system in which the various architectures and functionalities of the various previous implementations may be implemented. As shown, the computer system 500 includes at least one processor 504 that is connected to a bus 502, wherein the computer system 500 may be implemented using any suitable protocol, such as PCI (Peripheral Component Interconnect), PCI-Express, AGP (Accelerated Graphics Port), Hyper Transport, or any other bus or point-to-point communication protocol (s). The computer system 500 also includes a memory 506.
Control logic (software) and data are stored in the memory 506 which may take a form of random-access memory (RAM). In the disclosure, a single semiconductor platform may refer to a sole unitary semiconductor-based integrated circuit or chip. It should be noted that the term single semiconductor platform may also refer to multi-chip modules with increased connectivity which simulate on-chip modules with increased connectivity which simulate on-chip operation, and make substantial improvements over utilizing a conventional central processing unit (CPU) and bus implementation. Of course, the various modules may also be situated separately or in various combinations of semiconductor platforms per the desires of the user.
The computer system 500 may also include a secondary storage 510. The secondary storage 510 includes, for example, a hard disk drive and a removable storage drive, representing a floppy disk drive, a magnetic tape drive, a compact disk drive, digital versatile disk (DVD) drive, recording device, universal serial bus (USB) flash memory. The removable storage drive at least one of reads from and writes to a removable storage unit in a well-known manner.
Computer programs, or computer control logic algorithms, may be stored in at least one of the memories 506 and the secondary storage 510. Such computer programs, when executed, enable the computer system 500 to perform various functions as described in the foregoing. The memory 506, the secondary storage 510, and any other storage are possible examples of computer-readable media.
In an implementation, the architectures and functionalities depicted in the various previous figures may be implemented in the context of the processor 504, a graphics processor coupled to a communication interface 512, an integrated circuit (not shown) that is capable of at least a portion of the capabilities of both the processor 504 and a graphics processor, a chipset (namely, a group of integrated circuits designed to work and sold as a unit for performing related functions, and so forth).
Furthermore, the architectures and functionalities depicted in the various previous-described figures may be implemented in a context of a general computer system, a circuit board system, a game console system dedicated for entertainment purposes, an application-specific system. For example, the computer system 500 may take the form of a desktop computer, a laptop computer, a server, a workstation, a game console, an embedded system.
Furthermore, the computer system 500 may take the form of various other devices including, but not limited to a personal digital assistant (PDA) device, a mobile phone device, a smart phone, a television, and so forth. Additionally, although not shown, the computer system 500 may be coupled to a network (for example, a telecommunications network, a local area network (LAN), a wireless network, a wide area network (WAN) such as the Internet, a peer-to-peer network, a cable network, or the like) for communication purposes through an I/O interface 508.
It should be understood that the arrangement of components illustrated in the figures described is exemplary and that other arrangements may be possible. It should also be understood that the various system components (and means) defined by the claims, described below, and illustrated in the various block diagrams represent components in some systems configured according to the subject matter disclosed herein. For example, one or more of these system components (and means) may be realized, in whole or in part, by at least some of the components illustrated in the arrangements illustrated in the described figures.
In addition, while at least one of these components are implemented at least partially as an electronic hardware component, and therefore constitutes a machine, the other components may be implemented in software that when included in an execution environment constitutes a machine, hardware, or a combination of software and hardware.
Although the disclosure and its advantages have been described in detail, it should be understood that various changes, substitutions, and alterations can be made herein without departing from the spirit and scope of the disclosure as defined by the appended claims.
, Claims: We CLAIMS
1. A method for real-time enhancement of an X-ray image, the method comprising:
(a) receiving (300) a low-resolution X-ray image comprising noise from an X-ray imaging device (102);
(b) processing (302), via one or more processors (106), the low-resolution X-ray image with a unified deep learning model (114) configured to, in a single processing step, generate a high-resolution, enhanced image, wherein said model (114) comprises a sequence of integrated layers configured to:
(i) first, extract a feature map from the low-resolution X-ray image via a feature extraction layer (204);
(ii) second, refine the feature map via one or more intermediate layers including an attention layer (210) configured to apply an attention map to focus on diagnostically relevant features while suppressing noise; and
(iii) third, reconstruct the high-resolution, enhanced image from the refined and attention-modulated feature map via an upsampling layer (214);
(c) generating (304) an intermediate high-resolution, enhanced image; and
(d) providing (310) the high-resolution, enhanced image for display on a display (104).
2. The method of claim 1, wherein the attention layer (210) computes (302) the attention map using a convolution operation followed by a sigmoid activation function.
3. The method of claim 1, wherein the attention-modulated feature map is generated by performing (302) an element-wise multiplication of the refined feature map with the attention map.
4. The method of claim 1, wherein the one or more intermediate layers further comprise a shrinking layer (206), a mapping layer (208), and an expanding layer (212), arranged in sequence with the feature extraction layer (204), attention layer (210), and upsampling layer (214).
5. The method of claim 1, wherein the upsampling layer (214) reconstructs (302) the high-resolution, enhanced image using sub-pixel convolution.

6. The method of claim 1, wherein the unified deep learning model (114) is trained by minimizing a composite loss function that balances pixel-wise accuracy, noise suppression, and perceptual quality.
7. The method of claim 6, wherein the composite loss function comprises a pixel loss component, a perceptual loss component, and a pyramid loss component.
8. The method of claim 1, wherein the unified deep learning model (114) executed by the one or more processors (106) is integrated within the X-ray imaging device (102), such that the enhancement of the low-resolution X-ray image is performed onboard said X-ray imaging device (102).
9. A system (100) for real-time enhancement of an X-ray image, the system comprising:
(a) one or more processors (106); and
(b) a memory (108) storing instructions for a unified deep learning model (114) that, when executed by the one or more processors (106), cause the system (100) to perform the method of claim 1.
10. The system of claim 9, wherein the attention layer (210) of the unified deep learning model (114) is configured to compute the attention map using a convolution operation followed by a sigmoid activation function.
11. The system of claim 9, wherein the attention layer (210) of the unified deep learning model (114) is configured to generate the attention-modulated feature map by performing an element-wise multiplication of the refined feature map with the attention map.
12. The system of claim 9, wherein the one or more intermediate layers of the unified deep learning model (114) further comprise a shrinking layer (206), a mapping layer (208), and an expanding layer (212)
13. The system of claim 9, wherein the unified deep learning model (114) is trained by minimizing a composite loss function comprising a pixel loss component, a perceptual loss component, and a pyramid loss component.
14. The system of claim 9, wherein the method further comprises post-processing using a post-processing module (116) adapted to enable or disable post-processing operations to provide high-resolution, enhanced image for display.
15. The system of claim 9, wherein the upsampling layer (214) reconstructs the high-resolution, enhanced image using sub-pixel convolution.
16. The system of claim 9, wherein the one or more processors (106) and the memory (108) are integrated within an X-ray imaging device (102) such that the unified deep learning model (114) executes locally within said X-ray imaging device (102) to provide real-time enhancement of image data.