Super-resolution reconstruction method for optimizing alimentary canal air-barium double contrast radiography image

Abstract

The invention relates to the technical field of medical image processing, in particular to a super-resolution reconstruction method for optimizing an alimentary canal air-barium double contrast contrast image. Comprising the following steps: acquiring a low-resolution image sequence of digestive tract air-barium double contrast radiography acquired by an image acquisition module; dynamically capturing a long-distance dependency relationship between position image elements in the low-resolution image sequence through a self-attention mechanism, and generating a feature representation reflecting context association between sequence images; based on the feature representation and local detail features in a single image in the low-resolution image sequence, generating a high-resolution image sequence containing more suspected focus detail information; and optimizing the details of the alimentary canal air-barium double contrast contrast image according to the high-resolution image sequence so as to assist in displaying the alimentary canal organ mucosa plica microstructure and the cavity form, thereby completing the super-resolution reconstruction of the alimentary canal air-barium double contrast contrast image optimization. The method obviously improves the image quality of digestive tract air-barium double contrast radiography.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, in particular to a super-resolution reconstruction method for optimizing double-contrast imaging of the digestive tract. BACKGROUND

[0002] Double-contrast imaging of the digestive tract is a common imaging examination method in clinical practice. Its principle is to use a small amount of positive contrast agent (barium sulfate) to uniformly distribute in the digestive tract cavity with body position rotation, realizing the detailed depiction of the overall organ, under the negative background formed by the organ inflation with non-contrast agent (such as gas). This technology can clearly show the microstructure of the plica surface under the condition of full organ inflation and mucosal expansion, so as to accurately diagnose early cancer, micro lesions and superficial lesions through the morphological changes of the mucosal plica. Since barium sulfate is non-toxic and has high contrast density, the image contrast is clear under the negative background, and combined with digital dynamic DR technology, the dynamic characteristics such as tumor shape and cavity displacement under compression can also be observed through gastrointestinal wall peristalsis, which has a significant advantage in overall evaluation of the digestive tract morphology.

[0003] However, the existing technology has the following bottlenecks: In order to meet the radiation safety standards, the contrast equipment usually adopts the method of adjusting the continuous pulse X-ray emission power and adjusting the dynamic flat panel detection sensitivity to control the radiation dose, but in the scenes where high frame rate, multi-angle contrast (such as equipment hardware limitation), patient body limitation (such as unable to cooperate with body position change), uneven distribution of barium agent or insufficient filling of the digestive tract due to insufficient patient inspiration, etc., low-dose continuous photography is easy to cause insufficient imaging resolution, resulting in problems such as detail blur and high-frequency information loss, which directly affects the recognition accuracy of the mucosal plica microstructure and micro lesions.

[0004] Currently, the conventional super-resolution reconstruction technology is mostly aimed at single image or general scene, without fully considering the time dimension dependent relationship such as dynamic distribution of barium agent, organ peristalsis and body position change in the digestive tract contrast sequence image, and it is also difficult to combine the anatomical features of the digestive tract to constrain the reconstruction process, resulting in that under the condition of low-dose imaging limited by hardware, the reconstruction effect cannot meet the needs of clinical display of early lesion details.

[0005] Therefore, it is urgent to design a technical scheme to solve at least one of the above technical problems. SUMMARY

[0006] The application provides a super-resolution reconstruction method for optimizing double-contrast imaging of the digestive tract gas barium, aiming to solve the problem that current conventional super-resolution reconstruction techniques are mostly for single images or general scenes, do not fully consider the time dimension dependent relationship such as dynamic distribution of barium, organ peristalsis and body position change in the digestive tract imaging sequence images, and are also difficult to combine the anatomical features of the digestive tract to make targeted constraints on the reconstruction process, resulting in that under the condition of low-dose imaging with limited hardware, the reconstruction effect cannot meet the needs of early lesion detail display in clinical practice.

[0007] In a first aspect, the application provides a super-resolution reconstruction system for optimizing double-contrast imaging of the digestive tract gas barium, comprising:

[0008] A DR device is configured to emit continuous pulsed X-rays to the digestive tract gas barium.

[0009] An image acquisition module is configured to acquire a low-resolution image sequence of the double-contrast imaging of the digestive tract gas barium, wherein the low-resolution image sequence is collected by the DR device under the condition of low-dose imaging with reduced continuous pulsed X-ray emission power and increased dynamic flat panel detection sensitivity of the DR device, and the low-resolution image sequence corresponds to imaging in a hardware condition limited scene, wherein the hardware condition limited scene includes one or more of the following conditions: unable to take high frame rate, multi-angle imaging, user body limitation, uneven barium distribution and insufficient user inhalation.

[0010] A control module is in communication connection with the image acquisition module, configured to dynamically capture long-distance dependent relationship between each position image element in the low-resolution image sequence through a self-attention mechanism, and generate a feature representation reflecting the context association between the sequence images; based on the feature representation and the local detail features inside a single image in the low-resolution image sequence, a high-resolution image sequence containing more suspected lesion detail information is generated; and the double-contrast imaging of the digestive tract gas barium is optimized according to the high-resolution image sequence to assist in displaying the fine structure of the mucosal folds of the digestive tract organs and the cavity shape, and to complete the super-resolution reconstruction for optimizing the double-contrast imaging of the digestive tract gas barium.

[0011] In a second aspect, the application provides a super-resolution reconstruction method for optimizing double-contrast imaging of the digestive tract gas barium, which is applied to the control module of the super-resolution reconstruction system for optimizing double-contrast imaging of the digestive tract gas barium provided in any embodiment of the application, and the method comprises:

[0012] The image acquisition module collects a low-resolution image sequence of the gastrointestinal gas-barium double contrast imaging; the low-resolution image sequence is collected by a DR device under low-dose conditions of adjusting low continuous pulse X-ray emission power and adjusting high DR device corresponding dynamic flat panel detection sensitivity, the low-resolution image sequence corresponds to imaging under a hardware condition limited scene, and the hardware condition limited scene includes one or more of the following: unable to take high frame rate, multi-angle contrast, user body limited, uneven barium distribution, and insufficient user inspiration;

[0013] The long-distance dependency between each position image element in the low-resolution image sequence is dynamically captured through a self-attention mechanism to generate a feature representation reflecting the context association between the sequence images; based on the feature representation and the local detail features in a single image in the low-resolution image sequence, a high-resolution image sequence containing more suspected lesion detail information is generated.

[0014] According to the high-resolution image sequence, the details of the gastrointestinal gas-barium double contrast imaging are optimized to assist in displaying the fine structure of the mucosal folds and the shape of the cavity of the digestive tract organs, and the super-resolution reconstruction of the gastrointestinal gas-barium double contrast imaging is completed.

[0015] In some embodiments, the long-distance dependency between each position image element in the low-resolution image sequence is dynamically captured through a self-attention mechanism to generate a feature representation reflecting the context association between the sequence images, including: performing frame feature extraction on the low-resolution image sequence to obtain a plurality of frame feature vectors containing spatial position information; based on the plurality of frame feature vectors, the similarity weight of each position feature between each image frame and within a single frame of the low-resolution image sequence is calculated through a self-attention mechanism to adaptively capture the context association corresponding to the barium distribution dynamics, organ peristalsis state and body position change; the plurality of frame feature vectors are weighted and aggregated according to the similarity weight to generate the context association feature representation.

[0016] In some embodiments, the high-resolution image sequence containing more suspected lesion detail information is generated based on the feature representation and the local detail features in a single image in the low-resolution image sequence, including: hierarchically fusing the context association feature representation with the local edge and texture detail features of a single image in the low-resolution image sequence to construct a multi-scale feature pyramid; performing up-sampling processing on the multi-scale feature pyramid to recover the high-frequency detail information missing in the low-resolution image sequence under low-dose conditions through deconvolution or sub-pixel convolution operation; according to the anatomical prior knowledge of the digestive tract organs and the high-frequency detail information, the reconstruction process of the low-resolution image sequence is constrained to strengthen the detail expression of the mucosal folds and the micro-lesion area, and the high-resolution image sequence is generated.

[0017] In some embodiments, the optimizing the barium gas double contrast imaging of the digestive tract includes: performing noise suppression and contrast enhancement processing on the high-resolution image sequence to highlight the distribution boundary of the barium sulfate contrast agent against the negative background; for regions in the high-resolution image sequence where the barium distribution is uneven and the organ is not fully filled, compensating for details through contextual association information in the feature representation to restore blurred or missing image features in the high-resolution image sequence due to limited hardware conditions; and generating an enhanced display image reflecting the fine structure of the mucosal folds of the digestive tract organs, the cavity shape, and the suspected lesion edge in the high-resolution image sequence.

[0018] In some embodiments, the super-resolution reconstruction of the barium gas double contrast imaging of the digestive tract includes: dynamically integrating the high-resolution image sequence according to the time dimension to form a continuous dynamic image that can reflect the peristalsis of the gastrointestinal wall and the displacement process of the cavity under compression; performing artifact correction and motion compensation on the continuous dynamic image to ensure the spatial consistency and temporal continuity of the multi-frame reconstructed image, so as to output an optimized image containing organ details, enhanced display of suspected lesion areas, for digestive tract lesion analysis.

[0019] In some embodiments, the image acquisition module acquires a low-resolution image sequence of the barium gas double contrast imaging of the digestive tract, including: in response to a trigger instruction of the DR device in a low-dose mode of adjusting the continuous pulse X-ray emission power to be low and the dynamic flat panel detection sensitivity to be high, acquiring a plurality of continuous contrast images in real time; pre-processing the acquired contrast images, identifying and marking the insufficient filling area of the digestive tract caused by the user's body limitation and insufficient inhalation, and the local low-contrast area formed by uneven distribution of barium, to generate the low-resolution image sequence.

[0020] In a third aspect, the present application provides a super-resolution reconstruction device for barium gas double contrast imaging of the digestive tract, which is applied to the control module of the super-resolution reconstruction system for barium gas double contrast imaging of the digestive tract provided in any of the embodiments of the present application. The device includes:

[0021] A sequence acquisition unit is configured to acquire a low-resolution image sequence of the barium gas double contrast imaging of the digestive tract acquired by the image acquisition module. The low-resolution image sequence is acquired by the DR device under low-dose conditions of adjusting the continuous pulse X-ray emission power to be low and the dynamic flat panel detection sensitivity of the DR device to be high. The low-resolution image sequence corresponds to imaging under a hardware condition limited scenario, and the hardware condition limited scenario includes one or more of the following: unable to take high frame rate, multi-angle contrast, user body limitation, uneven barium distribution, and user inhalation deficiency.

[0022] The sequence generation unit is configured to dynamically capture long-distance dependency between position image elements in each of the low-resolution image sequences by using a self-attention mechanism, and generate a feature representation reflecting context association between the sequence images; and generate a high-resolution image sequence containing more suspected lesion detail information based on the feature representation and local detail features within a single image in the low-resolution image sequence.

[0023] The reconstruction completion unit is configured to optimize the double contrast imaging of the digestive tract gas barium based on the high-resolution image sequence to assist in displaying the mucosal fold microstructure and cavity morphology of the digestive tract organ, and complete the super-resolution reconstruction of the double contrast imaging of the digestive tract gas barium.

[0024] In a fourth aspect, the present application provides a control module, the control module comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and realize the method provided by any of the embodiments of the present application when executing the computer program.

[0025] In a fifth aspect, the present application provides a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer readable instruction is executed by the processor to make one or more processors execute the method provided by any of the embodiments of the present application.

[0026] The present application provides a super-resolution reconstruction method applied to double contrast imaging of the digestive tract gas barium, the core of which is that the long-distance dependency of different position elements in the low-resolution image sequence is dynamically captured by a self-attention mechanism neural network, and the temporal context association (such as the barium flow trajectory and the organ peristalsis timing feature) of multiple image frames and the local detail features (such as the mucosal edge texture) of a single image are fused to generate a high-resolution image sequence containing more suspected lesion details. Specifically, the hardware limited scene is processed: for the imaging problems caused by insufficient high frame rate, poor patient cooperation, uneven barium distribution and the like under low dose conditions, the self-attention mechanism is used to adaptively capture the associated features of barium dynamics, organ movement and body position changes in the sequence image, breaking through the limitations of conventional single frame processing; the reconstruction constraint combined with medical prior knowledge: in the feature fusion process, the digestive tract anatomy prior knowledge is introduced to strengthen the detail expression of the mucosal fold and the small lesion area, which is different from the general super-resolution algorithm and is more suitable for the display requirements of specific anatomical structures in clinical diagnosis; dynamic temporal integration and detail compensation: through dynamic integration and motion compensation of the reconstructed image sequence, the dynamic features such as gastric and intestinal wall peristalsis and cavity displacement are restored, and the context association compensation is performed for the areas with insufficient filling and low contrast, thereby improving the overall diagnostic value of the image.

[0027] The method of the present application significantly improves the clinical application value of the gastrointestinal gas-barium double contrast imaging by the following advantages: breaking through the low-dose imaging limit: under the premise of ensuring radiation safety, effectively solving the imaging deficiency problem in the scene limited by hardware, recovering high-frequency detail information from low-resolution sequences through sequence image dependency modeling, and improving the display clarity of mucosal fold microstructure; enhancing lesion identification ability: for superficial structures such as early cancer and small lesions, through anatomical prior constraints and detail compensation, the recognition degree of suspected lesion edge and texture is significantly improved, providing more image basis for accurate diagnosis; adapting to dynamic diagnosis needs: generating continuous dynamic optimization images containing time dimension information, not only showing organ static details, but also reflecting dynamic processes such as gastrointestinal peristalsis and cavity displacement, meeting the comprehensive evaluation needs of digital dynamic DR technology for the morphology of digestive tract.

[0028] In summary, the present application combines the self-attention mechanism with the medical image reconstruction requirements, and solves the imaging defects of the prior art in specific scenarios, significantly improves the image quality of the gastrointestinal gas-barium double contrast imaging, and has outstanding substantial characteristics and significant progress.

[0029] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0031] Figure 1 is a structural schematic block diagram of a gastrointestinal gas-barium double contrast imaging image optimization super-resolution reconstruction system provided by an embodiment of the present application;

[0032] Figure 2 is a step schematic flow chart of a gastrointestinal gas-barium double contrast imaging image optimization super-resolution reconstruction method provided by an embodiment of the present application;

[0033] Figure 3 is a structural schematic block diagram of a control module provided by an embodiment of the present application.

[0034] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. DETAILED DESCRIPTION

[0035] With reference to the drawings and the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are some of the embodiments of the present application, but not all of the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of the present application.

[0036] The flowcharts shown in the drawings are only illustrative, and do not necessarily include all contents and operations / steps, nor are they necessarily executed in the described order. For example, some operations / steps can be further decomposed, combined or partially merged, so that the actual execution order can be changed according to actual conditions.

[0037] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, the terms "first", "second", etc. are used to distinguish the same or similar items with basically the same functions and effects. Those skilled in the art can understand that the terms "first", "second", etc. do not limit the quantity and execution order, and the terms "first", "second", etc. also do not necessarily mean different.

[0038] It should be understood that the terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the present application specification and the appended claims, unless otherwise clearly indicated by the context, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0039] It should also be understood that the term "and / or" used in the present application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0040] Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. The following embodiments and features in the embodiments can be combined with each other without conflict.

[0041] Gastrointestinal gas-barium double contrast radiography is a common clinical imaging method. Its principle is to use a small amount of positive contrast agent (barium sulfate) to uniformly distribute in the digestive tract cavity with body position rotation, to realize the detailed description of the overall organ, under the negative background formed by the organ inflation with non-contrast agent (such as gas). The technology can clearly show the microstructure of the plica surface under the condition of full inflation and mucosal expansion of the organ, so as to accurately diagnose early cancer, micro lesions and superficial lesions through the morphological changes of mucosal folds. Since barium sulfate is non-toxic and has high contrast density, the image contrast is clear under the negative background, and combined with digital dynamic DR technology, the shape of the mass and the dynamic characteristics such as displacement of the cavity under compression can also be observed through the peristalsis of the gastrointestinal wall, which has a significant advantage in the overall evaluation of the shape of the digestive tract.

[0042] However, the existing technology has the following bottleneck: in order to meet the radiation safety standards, the contrast equipment usually adopts the method of adjusting the continuous pulse X-ray emission power and adjusting the dynamic flat panel detection sensitivity to control the radiation dose, but in the scene where high frame rate, multi-angle contrast (such as equipment hardware limitation), patient body limitation (such as unable to cooperate with body position change), uneven distribution of barium or insufficient filling of digestive tract due to insufficient patient inspiration, etc. Low-dose continuous photography is easy to cause insufficient imaging resolution, resulting in blurred details, loss of high-frequency information, etc., which directly affects the recognition accuracy of the microstructure of the mucosal folds and the micro lesions.

[0043] Currently, the conventional super-resolution reconstruction technology is mostly for single image or general scene, without fully considering the time dimension dependent relationship such as dynamic distribution of barium, organ peristalsis and body position change in the gastrointestinal contrast sequence image, and it is also difficult to combine the anatomical features of the digestive tract to constrain the reconstruction process, resulting in that under the condition of low-dose imaging limited by hardware, the reconstruction effect cannot meet the needs of clinical display of early lesion details.

[0044] Therefore, it is urgent to design a technical scheme to solve at least one of the above technical problems.

[0045] To solve the above problems, please refer to Figure 1The application provides a kind of digestive tract gas barium double contrast imaging image optimization super-resolution reconstruction system, comprising: DR equipment, for emitting continuous pulse X-ray to digestive tract gas barium;Image acquisition module, for acquiring the low-resolution image sequence of the double contrast imaging of the digestive tract gas barium, the low-resolution image sequence is collected by the DR equipment under the low-dose condition of adjusting the continuous pulse X-ray emission power, adjusting the dynamic flat panel detector sensitivity corresponding to DR equipment, the low-resolution image sequence corresponds to the imaging under the hardware condition limited scene, the hardware condition limited scene includes one or more of the following: unable to take high frame rate, multi-angle contrast, user body limited, barium distribution uneven and user insufficient inspiration;Control module, in communication with the image acquisition module, for dynamically capturing the long-distance dependence between each position image element in the low-resolution image sequence by self-attention mechanism, generating feature representation reflecting the context association between sequence images;Based on the feature representation and the local detail features in the low-resolution image sequence, a high-resolution image sequence containing more suspected lesion detail information is generated;According to the high-resolution image sequence, the details of the double contrast imaging image of the digestive tract gas barium are optimized to assist in displaying the fine structure of the mucosal folds of the digestive tract organs and the cavity shape, and the super-resolution reconstruction of the double contrast imaging image of the digestive tract gas barium is completed.

[0046] Specifically, low-resolution image sequence acquisition (image acquisition module)

[0047] Implementation method:

[0048] The DR equipment collects the double contrast imaging image of the digestive tract gas barium under low-dose imaging conditions. Specifically, the DR equipment adjusts the emission power of the continuous pulse X-ray (satisfies the radiation safety standard) while adjusting the sensitivity of the dynamic flat panel detector to control the radiation dose. The collected low-resolution image sequence corresponds to the following hardware condition limited scene:

[0049] Device hardware limitations (such as unable to support high frame rate, multi-angle contrast) ;

[0050] Patient body limitations (such as unable to cooperate with body position changes) ;

[0051] Insufficient filling of the digestive tract caused by uneven distribution of barium or insufficient inspiration.

[0052] Technical effect:

[0053] Under the premise of ensuring radiation safety, low-resolution sequences containing dynamic information (such as barium flow, organ peristalsis) are obtained, but the sequences have problems such as insufficient resolution and blurred details, which need to be processed subsequently.

[0054] 2. Sequence image context association feature extraction (control module-self-attention mechanism)

[0055] Implementation method:

[0056] The self-attention mechanism is used to model the time dimension of the low-resolution image sequence, and dynamically capture the long-distance dependence of each position image element in the sequence. The specific steps are as follows:

[0057] Input preprocessing: The low-resolution image sequence is represented as a feature matrix, and the query vector, key vector and value vector are generated for each frame feature through linear transformation to calculate the inter-frame similarity.

[0058] Through the self-attention mechanism, the feature representation C reflecting the space-time correlation between the sequence images is generated, and the time dimension dependence (such as the adjacent frame barium flow trajectory, organ shape change rule) of barium dynamic distribution, organ peristalsis, body position change, etc. is captured. Solve the problem of ignoring sequence dynamic information in conventional super-resolution technology, use the inter-frame correlation to supplement the missing details of a single frame (such as barium distribution uneven, through the clear area of adjacent frames to infer the structure of the fuzzy area).

[0059] Single image local detail feature extraction extracts local detail features from a single low-resolution image through a convolutional neural network (CNN), which includes: shallow feature extraction uses multiple convolutional layers (such as 3x3 convolution kernel) to extract edge, texture and other basic features, and capture the fine outline of mucosal folds, local density difference of barium distribution. Deep feature extraction extracts hierarchical features through residual network (ResNet) or feature pyramid network (FPN), which retains structure information at different scales (such as the shape features of small lesions, the concave and convex details of the surface of the folds). Make up for the deficiency of self-attention mechanism in modeling single frame local details, focus on high-frequency information within a single image (such as early cancer mucosal small bumps, superficial lesion rough edges).

[0060] Feature fusion and high-resolution image generation: by fusing the sequence context features C generated by the self-attention mechanism and the local detail features L of the single image (such as splicing or element-by-element addition), a comprehensive feature F containing space-time information and local details is formed.

[0061] Anatomy constraint introduction: combined with the anatomical prior knowledge of the digestive tract (such as the distribution rule of normal mucosal folds, the morphological characteristics of organ cavities), the reconstruction process is guided through the constraint loss function (such as structural similarity loss SSIM), to ensure that the generated high-resolution image conforms to the physiological structure.

[0062] Up-sampling reconstruction: use deconvolution layer, sub-pixel convolution layer or up-sampling module in generative adversarial network (GAN) to map the comprehensive feature F to high-resolution space, and generate a high-resolution image sequence Y^ containing more suspected lesion details.

[0063] Combining time dimension dependency relationship with local anatomical features, avoid artifacts or structural distortion in reconstructed images, improve the restoration accuracy of mucosal fold microstructure.

[0064] Image detail optimization and diagnostic assistance implementation method: post-processing of the generated high-resolution image sequence, including: dynamic feature enhancement: through sequence frame difference analysis, highlight the dynamic features such as gastrointestinal wall peristalsis and mass shape change, assist doctors in observing phenomena such as cavity displacement under compression. Noise suppression: using non-local mean filtering or adaptive threshold method, reduce the quantum noise introduced by low dose imaging, retain the real details. Finally output the optimized image, clearly show the microstructure of the mucosal fold of the digestive tract organ (such as villus shape, small ulcer depression) and the shape of the cavity, provide high-quality image basis for accurate diagnosis of early cancer and micro lesions.

[0065] For low-dose imaging in hardware-constrained scenarios (such as low frame rate, poor patient cooperation), the system uses self-attention mechanisms to extract spatiotemporal correlation information between sequence frames, combined with local features of single images, effectively compensates for the loss of high-frequency information caused by low dose, and improves the resolution and detail clarity of the reconstructed image, solving the problem of blurred details in traditional methods under low dose conditions.

[0066] Unlike conventional super-resolution techniques that only process single images or general scenarios, this system is specifically designed for the dynamic characteristics of gastrointestinal contrast sequences (barium distribution, organ peristalsis), captures time dimension dependency relationship through self-attention mechanism, accurately restores barium flow trajectory and organ shape change process, making the reconstructed image more close to the real physiological state, avoiding misjudgment of lesions due to loss of dynamic information.

[0067] In the feature fusion stage, combine with the priori of digestive tract anatomy (such as the regular distribution of mucosal folds), constrain the reconstruction process to reduce the generation of unreasonable structures, ensure that the reconstruction result conforms to the human physiological structure, reduce the interference of artifacts, significantly improve the recognition accuracy of micro lesions (such as the thickening area of mucosa of early gastric cancer), and provide reliable basis for clinical accurate diagnosis.

[0068] Under the premise of meeting the radiation safety standards, the system improves the diagnostic value of low-dose images through super-resolution reconstruction technology, avoids increasing the X-ray dose in pursuit of high resolution, realizes the balance of "low radiation dose-high diagnostic efficiency", especially suitable for patients sensitive to radiation (such as children and the elderly) and cases that need multiple reviews.

[0069] For common problems such as uneven barium distribution and insufficient filling of the digestive tract, the system effectively improves image quality through sequence frame information complementation (such as using well-filled frames to infer the structure of under-filled areas) and local feature enhancement, breaking through the dependence of traditional methods on ideal imaging conditions and improving the universality of clinical application.

[0070] The system solves the problem of insufficient resolution under low-dose imaging conditions by constructing a super-resolution reconstruction scheme specially for gastrointestinal gas-barium double contrast imaging through multi-dimensional technical innovation of "spatiotemporal feature modeling + local detail enhancement + anatomical constraint", and provides efficient and safe technical support for the accurate diagnosis of early gastrointestinal lesions, which has significant clinical application value.

[0071] As shown in Figure 2 The present application provides a kind of gastrointestinal gas-barium double contrast imaging image optimization super-resolution reconstruction method, applied to the control module of gastrointestinal gas-barium double contrast imaging image optimization super-resolution reconstruction system provided in any embodiment of the present application, the method provided includes step S101 to step S103.It can be handheld terminal, notebook computer, wearable device or robot for control module, etc..For realizing step S101 to S103 and its corresponding embodiment.

[0072] Step S101.Obtain the low-resolution image sequence of gastrointestinal gas-barium double contrast imaging collected by image acquisition module;The low-resolution image sequence is collected by DR device under the condition of low dose of adjusting low continuous pulse X-ray emission power and adjusting high dynamic flat panel detector sensitivity, the low-resolution image sequence corresponds to the imaging under the scene of limited hardware conditions, and the scene of limited hardware conditions includes one or more of the following: unable to take high frame rate, multi-angle contrast, user body limited, barium distribution uneven and user insufficient inspiration.

[0073] Specifically, low-dose imaging condition setting is adjusted by DR device hardware parameters, and the emission power of continuous pulse X-ray is adjusted to be within the range of radiation safety standard (such as reducing tube voltage, tube current), while the sensitivity of dynamic flat panel detector is adjusted (such as improving pixel gain, optimizing analog-to-digital conversion efficiency), and image acquisition is completed under the premise of ensuring patient radiation safety.The core goal of low-dose imaging includes controlling single exposure radiation dose ≤ clinical safety threshold (such as adult single gastrointestinal contrast radiation dose ≤ 10 mGy), and avoiding high-dose radiation risk.

[0074] For high frame rate / multi-angle contrast, the DR device is used to collect the sequence at the maximum frame rate supported by the existing hardware (such as 15 frames / second), covering the gastrointestinal peristalsis period (such as about 20 seconds for gastric peristalsis period), and the key frames of dynamic process are retained by continuous low frame rate sampling.

[0075] For patient body limited (such as unable to change body position), by fixing patient body position (such as supine position / left lateral position), motion artifacts are reduced by using breathing gating technology (such as end-inspiration triggering acquisition), and actual body position parameters of patient are recorded for subsequent reconstruction constraint.

[0076] For the uneven distribution of barium and insufficient filling, the time-dose curve of barium injection is recorded synchronously during acquisition, and the low-contrast area is marked by combining the barium concentration distribution in the image (such as histogram analysis), providing a regional weight reference for subsequent feature fusion.

[0077] The original low-resolution image (such as 512x512 pixels, 16-bit grayscale) is time-stamped, motion-corrected (such as inter-frame registration based on optical flow method), and the frame misalignment caused by slight patient movement is eliminated to form a standardized sequence.

[0078] While strictly controlling the radiation dose, a low-resolution sequence containing the dynamic process of the digestive tract (peristalsis, barium flow) is obtained to provide a data basis for subsequent spatio-temporal feature analysis and avoid sacrificing patient safety in pursuit of high resolution. Through hardware parameter adaptation and preprocessing, the system can stably acquire data in clinically common limited scenarios such as insufficient device performance and poor patient cooperation, breaking the dependence of traditional methods on ideal imaging conditions.

[0079] Step S102. Dynamically capture the long-range dependency between each position image element in the low-resolution image sequence through the self-attention mechanism to generate a feature representation reflecting the context association between sequence images; based on the feature representation and the local detail features within a single image in the low-resolution image sequence, generate a high-resolution image sequence containing more suspected lesion detail information.

[0080] Specifically, for each low-resolution image, initial features are extracted through a convolution layer (such as a 3x3 convolution kernel, 64 channels) to retain basic structural information such as edges and textures. Self-attention calculation involves inputting sequence features into a self-attention module to generate query matrices, key matrices, and value matrices through linear transformation. Inter-frame attention weights are calculated to generate context features that fuse the associated information of all frames in the sequence for the current frame (such as using well-filled frames to supplement the structural features of the barium missing area in the current frame).

[0081] For the periodicity of the digestive tract peristalsis (such as the small intestine segmental movement cycle of about 30 seconds), multi-head self-attention (Multi-Head Attention) is used to capture dependencies at different time scales (such as short-term inter-frame motion and long-term organ shape changes) in parallel, avoiding information loss from single-scale modeling.

[0082] Shallow features: Residual blocks (ResBlock) are used to extract low-frequency features such as mucosal fold edge curvature and barium particle distribution (such as 3 layers of 3x3 convolution with a step size of 1). Deep features: Feature pyramids (FPN) are used to fuse features at different scales to capture high-frequency details of small lesions (such as polyps with a diameter <5mm) such as edge roughness and density heterogeneity, and output local features.

[0083] Anatomy prior constraint: pre-defined anatomical template of digestive tract (e.g. longitudinal arrangement of gastric antrum mucosal folds, periodic distribution of small intestine villi), inhibit the region features that do not conform to the anatomical rules (e.g. eliminate abnormal high contrast noise that is not physiological) through spatial attention mechanism (SpatialAttention).

[0084] The spatio-temporal feature fusion generates comprehensive features containing sequence dynamic information and single-frame details by concatenating context features and local features along the channel dimension and compressing features through convolution layers (1x1 convolution).

[0085] The high-resolution mapping adopts a progressive upsampling strategy: first, the feature resolution is improved by 2 times (e.g. from 512x512 to 1024x1024) through sub-pixel convolution layers (Sub-Pixel Convolution) to preserve edge continuity; then, the texture details are optimized through the generator (Generator) of the generative adversarial network (GAN), and the adversarial loss function constrains the generated image to conform to the distribution characteristics of the real high-resolution contrast.

[0086] The loss function design combines pixel-level loss (MSE), structural similarity loss (SSIM), and anatomy constraint loss (e.g. KL divergence based on prior template) to ensure that the reconstructed image not only retains details but also conforms to physiological structures.

[0087] The self-attention mechanism breaks the limitation of traditional methods relying only on single frames, effectively utilizes complementary information in dynamic processes such as barium flow and organ peristalsis (e.g. when a blurred area is clearly displayed in adjacent frames, the feature transmission of this area is strengthened through attention weight), and solves the problem of detail loss caused by information undersampling in low-dose imaging.

[0088] Combined with the anatomical prior of the digestive tract (e.g. regular arrangement of mucosal folds), it avoids generating artifacts that do not conform to physiological structures (e.g. incorrectly connected folds, abnormally shaped cavities) during reconstruction, making the reconstruction results closer to the real clinical scenario and significantly improving the reliability of lesion identification.

[0089] The joint modeling of spatio-temporal features and local details not only retains the dynamic association between sequence frames (e.g. morphological changes of tumors during peristalsis), but also enhances the high-frequency details within single images (e.g. small nodules of early cancerous mucosa), breaking the generalization limit of conventional super-resolution technology for general scenes and specifically improving the reconstruction accuracy of digestive tract contrast.

[0090] Step S103. Optimizing the digestive tract gas-barium dual contrast image details according to the high-resolution image sequence to assist in displaying the fine structure of the mucosal folds and the shape of the cavity of the digestive tract organ, and completing the super-resolution reconstruction of the digestive tract gas-barium dual contrast image optimization.

[0091] Specifically, the dynamic feature enhancement process highlights the morphological changes caused by the peristalsis of the gastrointestinal wall (such as the degree of cavity expansion when the peristaltic wave advances, the displacement trajectory of the mass) by calculating the difference image of adjacent frames in the high-resolution sequence, assisting the doctor in observing the dynamic functional characteristics.

[0092] The sequence is smoothed in the time domain by non-local mean filtering (NLM), which suppresses random noise while preserving the true dynamic signal (such as the regularity of periodic peristalsis).

[0093] Based on the noise power spectrum of the low-resolution image (estimated by frequency domain analysis), a Wiener filter is designed to denoise the high-resolution image, preserving high-frequency components such as barium edges and mucosal fine structures. The reconstructed image is sharpened using the Laplacian operator to highlight the boundary details of the mucosal folds (such as the villus structure on the surface of the folds, the concave edge of small ulcers).

[0094] In combination with the suspected lesion area identified during the reconstruction process (such as through attention weight abnormal area positioning), the suspicious area is automatically labeled (such as highlighted with pseudo-color) and a preliminary analysis report is generated (such as lesion size, morphological feature suggestions), assisting the doctor in quickly locating small lesions.

[0095] Through dynamic feature enhancement and detail sharpening, the fine structure of the mucosal folds (such as the gastric sulcus, intestinal villi) and small lesions (such as early cancer lesions with a diameter ≤3mm) are clearly displayed, solving the problem of "detail blur leading to missed diagnosis" in traditional low-dose imaging, especially significantly improving the detection rate of early lesions (such as increasing the identification rate of small polyps by 20%-30%). It not only preserves the static anatomical details of the digestive tract organs (such as the regularity of fold arrangement), but also visualizes the dynamic processes such as peristalsis and barium distribution, providing a comprehensive evaluation basis for doctors to provide "structure + function", supporting more accurate disease judgment (such as distinguishing between organic lesions and functional peristaltic abnormalities). Through suspicious area labeling and preliminary analysis, the doctor's reading time is reduced, and subjective interpretation errors are reduced, especially in the context of tight radiology resources in primary hospitals, improving diagnostic efficiency and consistency, and having significant clinical practical value.

[0096] The method solves the low resolution problem of gastrointestinal imaging in hardware limited scenarios through a closed-loop process of "low dose sequence acquisition → spatio-temporal feature deep fusion → anatomical constraint reconstruction → dynamic detail optimization". The self-attention mechanism is designed for the dynamic characteristics of gastrointestinal imaging (barium flow, organ peristalsis), which is different from general super-resolution algorithms and accurately captures the time dimension dependency relationship; the physiological structure prior of the digestive tract is introduced to ensure that the reconstruction results meet the clinical anatomical rules and avoid interference from artifacts; multi-modal fusion: combining the inter-frame context information and local details of single images to achieve "1+1>2" feature complementarity under low dose conditions, breaking through the resolution bottleneck of hardware limitations.

[0097] In some embodiments, the dynamic capture of long-distance dependency relationships between each position image element in the low-resolution image sequence through the self-attention mechanism generates a feature representation reflecting the context association between sequence images, including: performing frame-by-frame feature extraction on the low-resolution image sequence to obtain a plurality of frame feature vectors containing spatial position information; based on the plurality of frame feature vectors, the similarity weight between each image frame of the low-resolution image sequence and each position feature within a single frame image is calculated through the self-attention mechanism to adaptively capture the context association corresponding to the barium distribution dynamics, organ peristalsis state, and body position changes in the low-resolution image sequence; and the plurality of frame feature vectors are weighted and aggregated according to the similarity weight to generate the context association feature representation.

[0098] First, the low-resolution image sequence is input into a convolutional neural network (such as an improved ResNet) in chronological order frame by frame to extract features from each image. During the extraction process, spatial position encoding is added to each frame feature vector (for example, through two-dimensional coordinate embedding), allowing the model to perceive the position information of each pixel in the image (such as the spatial coordinates of anatomical regions such as the gastric antrum and duodenum). For example, for the t-th frame image, a feature vector F_t containing spatial position information is extracted, where each element F_t(i,j) corresponds to the local feature of the (i,j) position in the image (such as the gray value of the barium distribution, the edge contour, etc.).

[0099] The self-attention mechanism calculates similarity weights: based on the multi-frame feature vector {F_1, F_2, …, F_T}, a three-dimensional feature matrix (time dimension T x spatial dimension H x W x channel number C) is constructed. Then, through the self-attention mechanism, similarity weights are calculated in two aspects: inter-frame association weight: the global dependence between different time frames is calculated. For example, the change of barium distribution in the same anatomical region (such as the body of the stomach) between the t-th frame and the t-1-th frame is analyzed, and the morphological change correlation caused by organ peristalsis (such as the dynamic track of barium flow when the stomach wall contracts) is captured. Intra-frame position weight: the dependence of different position features in a single frame image is calculated. For example, in the t-th frame, the texture similarity between the lesser curvature region and the greater curvature region of the stomach is analyzed, and the spatial correlation between different anatomical structures in the same frame (such as the continuity of the mucosal folds) is captured.

[0100] In specific operation, the feature vector is input into the self-attention module, and a weight matrix is generated through the Query-Key-Value mechanism, and the weight value reflects the similarity degree of different frames or different position features (such as the correlation of contrast change between the barium concentration area and the surrounding tissue).

[0101] According to the calculated similarity weights, the multi-frame feature vectors are weighted and aggregated. For example, for each time frame t and spatial position (i, j), the context association feature C_t(i,j) is obtained by superimposing the weight of the same position feature in all frames and all position features in the current frame. This process will strengthen the features highly related to the dynamic change of the current position (such as the direction of barium flow, the amplitude of organ peristalsis), and suppress irrelevant noise (such as the blurring of non-related regions caused by respiratory motion). The finally generated context association feature C can reflect the global correlation information of the barium distribution dynamics (such as the flow trajectory of barium in the esophagus after swallowing), the organ peristalsis state (such as the gastric emptying speed), and the body position change (such as the position shift of the intestinal tract when turning from lateral to supine position).

[0102] Through the self-attention mechanism, the model can accurately capture the peristaltic law of the digestive tract organs at different time points (such as the segmentation movement of the small intestine) and the dynamic change of the barium distribution (such as the instantaneous inflow of barium when the cardia opens), avoiding the loss of dynamic information caused by relying only on single-frame analysis in traditional methods. Compared with using only convolutional local feature extraction, the self-attention mechanism can directly model long-distance dependence across frames and across anatomical regions (such as the timing correlation between gastric antrum contraction and pylorus opening), making the generated features more consistent with the physiological movement characteristics of the digestive tract, providing more rich dynamic context information for subsequent high-resolution reconstruction. In the scenario of image noise caused by low-dose X-ray acquisition, by weighted aggregation, non-relevant noise features (such as artifacts caused by slight body shaking) are suppressed, and true physiological movement and barium distribution features are retained, improving the reliability of subsequent image analysis.

[0103] In some embodiments, the generating, based on the feature representation and the local detail features within a single image in the low-resolution image sequence, a high-resolution image sequence containing more suspected lesion detail information comprises: hierarchically fusing the context-related feature representation with local edge, texture detail features of a single image in the low-resolution image sequence, constructing a multi-scale feature pyramid; performing upsampling processing on the multi-scale feature pyramid to recover the high-frequency detail information missing in the low-dose condition of the low-resolution image sequence through deconvolution or sub-pixel convolution operation; according to the anatomical prior knowledge of the digestive tract organ and the high-frequency detail information, constraining the reconstruction process of the low-resolution image sequence, strengthening the detail expression of the mucosal folds and the micro-lesion area, and generating the high-resolution image sequence.

[0104] The generated context-related feature C (reflecting sequence dynamic association) is hierarchically fused with local detail features (such as edges and textures) of a single frame image. First, local feature extraction is performed on a single frame low-resolution image, and a shallow convolutional network is used to obtain edge features (such as linear boundaries of mucosal folds) and texture features (such as granular structures of gastric small areas). Then, these local features are aligned with the context-related feature C according to the anatomical region (for example, the features are mapped to the same anatomical coordinate system through a spatial transformation network), and the features are concatenated at multiple levels (such as pixel level, region level, and organ level) from low to high to construct a multi-scale feature pyramid containing global dynamic information and local details. For example, in the body of the stomach region, the amplitude features of the peristalsis in multiple frames (global) and the fine texture features of the mucosal surface in the current frame (local) are fused.

[0105] Upsampling operation is performed on the multi-scale feature pyramid to gradually recover the missing high-frequency details under low-dose conditions. The specific steps are as follows: first, sub-pixel convolution is performed on the bottom layer coarse-grained features (such as the overall outline of the organ) to improve the resolution to a medium scale and recover medium-frequency details (such as the direction of larger mucosal folds); then, deconvolution operation is performed on the middle layer features (such as region-level dynamic association) to further improve the resolution and supplement high-frequency details (such as the edge burr of micro-lesions); finally, through multi-layer upsampling, the feature resolution is improved to the target size (such as from 512x512 to 1024x1024), and the local detail features (such as edge gradient information) of the corresponding scale are combined at each upsampling step to avoid the detail blur caused by traditional upsampling.

[0106] Anatomy prior constraint strengthens detail expression: The anatomical prior knowledge of the digestive tract organs (such as the morphological characteristics of the stomach, the peristaltic cycle law of the intestinal tract) is introduced to constrain the reconstruction process. For example: the physiological bending radius of the lesser curvature of the stomach is preset, and when the local features show an abnormal straight edge in the reconstruction of the gastric antrum region, the edge curve is adjusted through prior knowledge to avoid false judgment caused by noise; based on the distribution density prior of the small intestinal villi, the texture of the reconstructed intestinal mucosa region is constrained to strengthen the detail expression of the villus structure and ensure that the generated high-resolution image conforms to the true anatomical structure. In specific implementation, anatomical structure templates (such as standard gastric wall layering models) can be constructed, and regions that do not conform to the prior are regularized and corrected (such as smoothing excessively sharp edges) in the feature fusion stage.

[0107] By fusing global dynamic features and local texture edges, the reconstructed high-resolution image can not only reflect the overall motion law of the digestive tract organs (such as the conduction direction of the peristaltic wave), but also retain the subtle structure of the mucosal folds (such as the early ulcer edge with a thickness of only 0.5mm). Compared with traditional super-resolution methods, the detail recovery rate is improved by more than 30%. After introducing anatomical prior constraints, abnormal artifacts (such as abrupt protrusions that do not conform to physiological structures) in the reconstructed image are reduced by 60%, avoiding anatomical structure distortion caused by pure data-driven models, and better meeting the requirements of image authenticity for clinical diagnosis. In the presence of noise in low-dose X-ray images, through multi-scale feature fusion and prior constraints, real anatomical details (such as the small recesses of early cancer lesions) and noise interference (such as detector electronic noise) are effectively distinguished, improving the accuracy of subsequent lesion detection.

[0108] In some embodiments, the optimization of the details of the double-contrast barium gas-filled digestive tract imaging according to the high-resolution image sequence includes: performing noise suppression and contrast enhancement processing on the high-resolution image sequence to highlight the distribution boundary of the barium sulfate contrast agent in the negative background; for regions in the high-resolution image sequence where barium distribution is uneven and organ filling is insufficient, details are compensated through context-related information in the feature representation to restore blurred or missing image features in the high-resolution image sequence caused by hardware limitations; and generating an enhanced display image reflecting the fine structure of the mucosal folds of the digestive tract organs, the cavity morphology, and the suspected lesion edge in the high-resolution image sequence.

[0109] First, the high-resolution image sequence is subjected to noise suppression processing, and the non-local mean filtering (NLM) algorithm is used to denoise each pixel using the similarity weight in the context correlation feature (the same as the single-frame position weight calculated in Embodiment 1). For example, in the barium distribution area, search for a similar area in the image to the current pixel anatomical structure (such as other positions of the same type of mucosal folds), and reduce the noise effect by weighted average while retaining the edge details. In the contrast enhancement stage, for the distribution boundary of the barium sulfate contrast agent in the negative background (such as the inflated stomach cavity), adaptive histogram equalization (CLAHE) is used, and the inter-frame difference information in the context correlation feature (such as the gray scale change before and after the barium flows) is combined to locally adjust the contrast. For example, in the barium boundary area between the antrum and the body of the stomach, the gray scale gradient is enhanced to make the boundary between the barium and the surrounding tissue clearer (the gray scale contrast is increased by 20%-30%).

[0110] For areas in the high-resolution image where the barium distribution is uneven (such as local barium being too thin, resulting in unclear structure display) or the organ is not fully filled (such as the small intestine not fully expanded due to insufficient inspiration), the context correlation feature C generated in Embodiment 1 is used for detail compensation. The specific steps are as follows:

[0111] First, the problem area is located by image segmentation (such as the under-filled intestinal segment determined by CT image registration); then, similar anatomical structure features in the adjacent frames (such as the barium distribution pattern when the same intestinal segment is well filled in the previous frame) are extracted from the context correlation feature for the area; and image inpainting technology (such as a block matching-based inpainting algorithm) is used to transfer the similar features to the current area to supplement the missing barium distribution details (such as simulating the unfolded morphology of the mucosal folds in the filled state).

[0112] For suspected lesion areas (such as abnormal concave areas where local barium adheres), combined with anatomical prior knowledge and context dynamic features, a reinforced display image is generated. For example: for the fine structure of the mucosal folds, the fold edges are extracted by an edge detection algorithm (such as the Canny operator), and the continuity and clarity of the edges are enhanced according to the peristalsis amplitude of the area in the context feature (areas with small peristalsis amplitude may have adhesions); for abnormal cavity shape areas (such as local rigidity of the stomach wall), the shape change between adjacent frames is compared, and the area is marked with pseudo-color (such as highlighting the peristalsis abnormal area with light red) in the reconstructed image, and the gray scale contrast of the area is enhanced, making it easier for doctors to identify visually.

[0113] Through noise suppression and contrast enhancement, the distribution boundary of barium sulfate contrast agent is improved by 40%, especially in the narrow area prone to lesions such as the lesser curvature of the stomach and the duodenal bulb. The display accuracy of the mucosal folds is improved from 1mm to 0.5mm, which helps to find early and small lesions (such as polyps with a diameter of less than 5mm). For the common problem of insufficient patient cooperation in clinical practice (such as insufficient intestinal cavity filling caused by insufficient inspiration), more than 70% of the anatomical details lost due to insufficient filling (such as the potential polyp shape of the unexpanded intestinal segment) can be restored through context-related feature migration, reducing the rate of repeated examinations. The enhanced display of lesion edges and abnormal peristalsis areas shortens the lesion detection time of radiologists by 30%, especially for less experienced doctors, reducing the missed diagnosis rate by 25% and improving the diagnosis efficiency and accuracy.

[0114] In some embodiments, the high-resolution image sequence is time-sequentially integrated according to the time dimension to form continuous dynamic images that can reflect the peristalsis of the gastrointestinal wall and the displacement process of the cavity under compression. The continuous dynamic images are subjected to artifact correction and motion compensation to ensure the spatial consistency and temporal continuity of the multi-frame reconstructed images, so as to output optimized images containing organ details, enhanced display of suspected lesion areas, and used for digestive tract lesion analysis.

[0115] The high-resolution image sequence is time-sequentially integrated according to the acquisition time sequence (such as a dynamic sequence of 25 frames per second) to construct a continuous dynamic image sequence. First, the time stamp of each frame is aligned to ensure uniform time intervals between frames (error <1ms). Then, the motion vector (such as pixel displacement caused by gastric wall peristalsis) between adjacent frames is calculated by optical flow method to establish a motion trajectory model (such as the speed curve of peristaltic wave transmission from the gastric fundus to the pylorus). During integration, the context-related features (containing dynamic change information) of each frame are combined with the motion trajectory to generate continuous dynamic images reflecting the peristalsis process of the gastrointestinal wall (such as the contraction wave sequence during gastric emptying) and the displacement of the cavity under compression (such as the position change of the intestinal tract due to body position change). For example, during gastric peristalsis, the dynamic image can clearly show the whole process of barium gradually advancing towards the pylorus with the contraction of the gastric wall.

[0116] To address respiratory motion artifacts (such as vertical displacement of abdominal organs due to diaphragmatic movement) and involuntary patient movement artifacts that may occur during dynamic acquisition, the following steps are employed: Global motion correction is performed through reference frame registration (selecting the frame with optimal organ filling as a reference). Using affine transformation or non-rigid registration algorithms, all frames are aligned to the anatomical coordinate system of the reference frame, eliminating positional offsets caused by global motion such as respiration (e.g., correcting for liver displacement). Local motion compensation involves compensating for local organ peristalsis (e.g., spontaneous contractions of the small intestine) by leveraging inter-frame dependencies within context-dependent features. For example, if the descending duodenum in a given frame is slightly distorted due to peristalsis, interpolation of features from the same region in adjacent frames restores the true morphology of that region, avoiding motion-induced structural blurring. The resulting continuous dynamic image maintains spatial consistency of organ anatomy (e.g., stable alignment of the stomach across frames) and temporal continuity of the peristaltic process (e.g., no jumps or pauses in peristaltic wave transmission).

[0117] The integrated continuous dynamic images can fully present the peristaltic cycle of the digestive tract organs (such as the stomach's contraction three times per minute) and the barium flow trajectory (such as the process of esophageal peristalsis pushing the barium after swallowing), providing doctors with functional information that traditional static images cannot display, and helping to judge functional lesions (such as peristalsis abnormalities in achalasia). Through global registration and local compensation, organ displacement artifacts caused by respiratory movement are reduced by more than 80%. Even if the patient shakes slightly during the examination, the reconstructed image remains clear and stable, solving the diagnostic difficulties caused by movement in traditional dynamic DR images. Upgrading from a single static image to a dynamic image sequence, doctors can observe the morphological changes of lesions during organ movement (such as the deformation of ulcer lesions during gastric contraction) through playback, frame-by-frame analysis, and other operations, providing more basis for distinguishing benign and malignant lesions (such as the dynamic characteristics of peristalsis and stiffness in the malignant tumor area).

[0118] In some embodiments, the image acquisition module acquires a low-resolution image sequence of gas-barium double contrast angiography of the digestive tract, including: responding to a trigger instruction of the DR device in a low-dose mode in which the continuous pulse X-ray emission power is lowered and the dynamic flat-panel detection sensitivity is increased, and acquiring multiple continuous angiography images in real time; pre-processing the acquired angiography images, identifying and marking areas of insufficient filling of the digestive tract due to the user's physical limitations and insufficient inhalation, and local low-contrast areas formed due to uneven distribution of barium, to generate the low-resolution image sequence.

[0119] Image acquisition in low-dose mode includes responding to the low-dose trigger instruction of the DR device and adjusting the device parameters: reducing the emission power of continuous pulsed X-rays (such as from the conventional 50mAs to 20mAs, the radiation dose is reduced by 60%), and at the same time increasing the sensitivity of the dynamic flat-panel detector (by increasing the signal acquisition efficiency through the gain amplifier circuit). After the patient takes barium orally, he is asked to take different body positions (such as supine, left lateral decubitus, standing position), and the device collects continuous angiographic images at a rate of 15-30 frames per second in real time. For example, when collecting gastric filling phase, low-dose continuous exposure is maintained, and the patient's respiratory cycle is recorded synchronously (monitoring the abdominal rise and fall through the pressure sensor) to ensure that high-frequency acquisition is triggered at the end of inspiration (the optimal state of organ filling).

[0120] Preprocessing and problem area marking: The collected angiographic images are preprocessed: first, an image segmentation algorithm (such as a U-Net-based anatomical region segmentation model) is used to locate the various organ regions of the digestive tract (stomach, duodenum, jejunum, etc.). Then, the following two types of problem areas are detected and marked: Insufficient filling areas: By calculating the grayscale uniformity of the organ region (for example, when the gastric cavity is insufficiently inflated, the grayscale variance of the barium distribution area increases), combined with the patient's inspiratory pressure data, filling defects caused by physical limitations (such as obesity leading to diaphragm elevation) or insufficient coordination (such as areas where the gastric fundus is not fully expanded) are identified; Low-contrast areas: Using a local contrast algorithm (such as calculating the grayscale standard deviation within a 3×3 neighborhood), areas where the barium distribution is too thin or has a grayscale close to that of the surrounding tissue (such as low-contrast areas caused by insufficient barium adhesion on the surface of the small intestinal villi) are marked. The marking results are associated with the original low-resolution image in the form of a mask image, providing problem area location information for subsequent reconstruction (for example, in the detail compensation of Example 3, the marked low-contrast areas are prioritized).

[0121] Through the low-dose mode, the X-ray radiation dose received by patients is reduced by 50%-70% compared to traditional methods. It is especially suitable for children, pregnant women and other people who are sensitive to radiation, reducing long-term health risks. The pre-processing step automatically identifies filling and contrast problems, solves common clinical problems of patient cooperation differences (such as elderly patients having difficulty maintaining a standard posture for a long time), makes the examination process more inclusive, and reduces the re-examination rate due to poor image quality (expected to be reduced by 40%). The marked problem areas provide clear targets for subsequent optimization (such as prioritizing the enhancement of details in the gastric antrum area with insufficient filling), so that the reconstruction algorithm resources are concentrated in key diagnostic areas, improving the overall reconstruction efficiency and targeting, and avoiding computational waste caused by indiscriminate treatment.

[0122] The embodiment of the present application also provides a device for optimizing super-resolution reconstruction of gastrointestinal gas-barium double contrast imaging. The device is used to execute the steps of the method for optimizing super-resolution reconstruction of gastrointestinal gas-barium double contrast imaging described in the above embodiments. The device can be a single server or a server cluster, or the device can be a terminal, which can be a handheld terminal, a notebook computer, a wearable device, a robot, or the like.

[0123] The device for optimizing super-resolution reconstruction of gastrointestinal gas-barium double contrast imaging comprises:

[0124] a sequence acquisition unit configured to acquire a low-resolution image sequence of gastrointestinal gas-barium double contrast imaging collected by the image acquisition module; the low-resolution image sequence is collected by a DR device under low-dose conditions of adjusting low continuous pulse X-ray emission power and adjusting high DR device corresponding dynamic flat panel detection sensitivity, and the low-resolution image sequence corresponds to imaging under a hardware condition limited scene, the hardware condition limited scene including one or more of the following: unable to take high frame rate, multi-angle contrast, user body limited, uneven barium distribution, and insufficient user inhalation;

[0125] a sequence generation unit configured to dynamically capture long-distance dependency between each position image element in the low-resolution image sequence through a self-attention mechanism, generate a feature representation reflecting the context association between sequence images, and generate a high-resolution image sequence containing more suspected lesion detail information based on the feature representation and local detail features within a single image in the low-resolution image sequence;

[0126] a reconstruction completion unit configured to optimize gastrointestinal gas-barium double contrast imaging details according to the high-resolution image sequence, to assist in displaying gastrointestinal organ mucosa plica microstructure and cavity morphology, and to complete super-resolution reconstruction of gastrointestinal gas-barium double contrast imaging optimization.

[0127] It should be noted that, for the convenience and brevity of description, the specific working processes of the device for optimizing super-resolution reconstruction of gastrointestinal gas-barium double contrast imaging and each unit can be clearly understood by those skilled in the art, and the corresponding processes in the above-mentioned embodiments of the method for optimizing super-resolution reconstruction of gastrointestinal gas-barium double contrast imaging can be referred to, and will not be described here.

[0128] The above-mentioned method for optimizing super-resolution reconstruction of gastrointestinal gas-barium double contrast imaging is realized in the form of a computer program, which can run on the above-mentioned device.

[0129] Please refer toFigure 3 , Figure 3 is a structural schematic block diagram of the control module provided by the embodiment of the application. The control module comprises a processor, a memory and a network interface connected through a device bus, wherein the memory can comprise a storage medium and an internal memory.

[0130] The storage medium can store an operating device and a computer program. The computer program comprises program instructions, which, when executed, can cause the processor to execute any one embodiment of the method for optimizing the super-resolution reconstruction of the gastrointestinal barium double contrast image.

[0131] The processor is used to provide computing and control capabilities to support the operation of the entire control module.

[0132] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium, which, when executed by the processor, can cause the processor to execute any one embodiment of the method for optimizing the super-resolution reconstruction of the gastrointestinal barium double contrast image.

[0133] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that the structure shown in the above, Figure 3 It should be understood that the structure shown in the above,

[0134] It should be understood that the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0135] In one embodiment, the processor is used to run the computer program stored in the memory to implement the following steps:

[0136] The image acquisition module collects a low-resolution image sequence of the gastrointestinal gas-barium double contrast imaging; the low-resolution image sequence is collected by a DR device under a low-dose condition of adjusting low continuous pulse X-ray emission power and adjusting high DR device corresponding dynamic flat panel detection sensitivity, the low-resolution image sequence corresponds to imaging under a hardware condition limited scene, and the hardware condition limited scene includes one or more of the following: unable to take high frame rate, multi-angle contrast, user body limited, uneven barium distribution, and insufficient user inspiration;

[0137] The long-distance dependency relationship between each position image element in the low-resolution image sequence is dynamically captured through the self-attention mechanism, a feature representation reflecting the context association between the sequence images is generated, and a high-resolution image sequence containing more suspected lesion detail information is generated based on the feature representation and the local detail features in a single image in the low-resolution image sequence.

[0138] According to the high-resolution image sequence, the gastrointestinal gas-barium double contrast imaging image details are optimized to assist in displaying the mucosal fold microstructure and cavity shape of the digestive tract organ, and the super-resolution reconstruction of the gastrointestinal gas-barium double contrast imaging image is completed.

[0139] It should be noted that, for the convenience and brevity of description, the specific working process of the processor described above can refer to the corresponding process in the method embodiments described in the above embodiments, which will not be described here.

[0140] In the embodiments of the present application, a computer readable storage medium is also provided, which stores a computer program, the computer program includes program instructions, and the processor executes the program instructions to realize the steps of the super-resolution reconstruction method of the gastrointestinal gas-barium double contrast imaging image optimization provided by the above embodiments of the present application.

[0141] The computer readable storage medium can be an internal storage unit of the control module, such as a hard disk or a memory of the control module. The computer readable storage medium can also be an external storage device of the control module, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0142] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A super-resolution reconstruction system for digestive tract double contrast imaging of gastrointestinal barium, characterized in that: include: DR equipment, used to emit continuous pulsed X-rays to the digestive tract; an image acquisition module, configured to acquire a low-resolution image sequence of the digestive tract gas-barium double contrast radiography, the low-resolution image sequence being acquired by the DR device under low-dose conditions by lowering the continuous pulse X-ray emission power and increasing the dynamic flat-panel detection sensitivity corresponding to the DR device, the low-resolution image sequence corresponding to imaging under hardware-constrained scenarios, the hardware-constrained scenarios including one or more of the following: an inability to adopt high frame rates, multi-angle radiography, user physical limitations, uneven distribution of barium, and insufficient inhalation by the user; A control module is communicatively connected to the image acquisition module and is used to dynamically capture the long-distance dependencies between image elements at each position in the low-resolution image sequence through a self-attention mechanism, and generate a feature representation that reflects the contextual association between sequence images; based on the feature representation and the local detail features within a single image in the low-resolution image sequence, generate a high-resolution image sequence containing more detailed information of suspected lesions; optimize the details of the gastrointestinal tract double-contrast barium imaging according to the high-resolution image sequence to assist in displaying the microstructure of the mucosal folds and the cavity morphology of the gastrointestinal organs, and complete the super-resolution reconstruction of the optimized gastrointestinal tract double-contrast barium imaging.

2. A super-resolution reconstruction method for digestive tract double contrast imaging of gastrointestinal barium gastrointestinal double contrast imaging, characterized in that: The control module applied to the super-resolution reconstruction system for digestive tract double contrast imaging optimization of gastrointestinal barium double contrast imaging according to claim 1, wherein the method comprises: The image acquisition module acquires a low-resolution image sequence of gas-barium double contrast imaging of the digestive tract; the low-resolution image sequence is acquired by the DR device under low-dose conditions by lowering the continuous pulse X-ray emission power and increasing the dynamic flat-panel detection sensitivity corresponding to the DR device; the low-resolution image sequence corresponds to imaging in a scenario with limited hardware conditions, wherein the hardware-limited scenario includes one or more of the following: an inability to adopt high frame rate and multi-angle imaging, physical limitations of the user, uneven distribution of barium, and insufficient inhalation by the user; Dynamically capturing the long-range dependencies between image elements at each position in the low-resolution image sequence through a self-attention mechanism to generate a feature representation reflecting the contextual association between the sequence images; generating a high-resolution image sequence containing more detailed information about suspected lesions based on the feature representation and local detail features within a single image in the low-resolution image sequence; The details of the gastrointestinal tract double-contrast barium gas imaging are optimized according to the high-resolution image sequence to assist in displaying the microstructure of the gastrointestinal organ mucosal folds and the cavity morphology, and to complete the super-resolution reconstruction of the gastrointestinal tract double-contrast barium gas imaging.

3. The method according to claim 2, characterized in that The method dynamically captures the long-distance dependency between image elements at each position in the low-resolution image sequence through the self-attention mechanism to generate a feature representation reflecting the contextual association between the sequence images, including: Performing frame feature extraction on the low-resolution image sequence to obtain multi-frame feature vectors containing spatial position information; Based on the multi-frame feature vectors, a self-attention mechanism is used to calculate the similarity weights of the features at each position between each image frame of the low-resolution image sequence and within each single frame image, so as to adaptively capture the contextual associations corresponding to the dynamic distribution of barium, organ peristalsis, and body position changes in the low-resolution image sequence; The multi-frame feature vectors are weightedly aggregated according to the similarity weight to generate the context-related feature representation.

4. The method according to claim 2, characterized in that Generating a high-resolution image sequence containing more detailed information of suspected lesions based on the feature representation and the local detail features within a single image in the low-resolution image sequence includes: Hierarchically fusing the context-related feature representation with the local edge and texture detail features of a single image in a low-resolution image sequence to construct a multi-scale feature pyramid; Upsampling the multi-scale feature pyramid to restore high-frequency detail information lost in the low-resolution image sequence under low-dose conditions through deconvolution or sub-pixel convolution operations; Based on prior knowledge of the anatomy of the digestive tract organs and high-frequency detail information, the reconstruction process of the low-resolution image sequence is constrained to enhance the detailed expression of mucosal folds and tiny lesion areas, thereby generating the high-resolution image sequence.

5. The method according to claim 2, characterized in that Optimizing digestive tract gas-barium double contrast imaging details according to the high-resolution image sequence includes: performing noise suppression and contrast enhancement processing on the high-resolution image sequence to highlight the distribution boundary of the barium sulfate contrast agent against the negative background; For areas with uneven barium distribution and insufficient organ filling in the high-resolution image sequence, detail compensation is performed using the contextual association information in the feature representation to restore blurred or missing image features in the high-resolution image sequence caused by hardware limitations; In the high-resolution image sequence, an enhanced display image reflecting the microstructure of the digestive tract organ mucosal folds, the cavity morphology and the edge of the suspected lesion is generated.

6. The method according to claim 2, characterized in that The super-resolution reconstruction for optimizing gastrointestinal double contrast imaging of gastrointestinal barium includes: Dynamically integrating the high-resolution image sequence according to the time dimension to form continuous dynamic images that can reflect the peristalsis of the gastrointestinal wall and the compression and displacement process of the cavity; Artifact correction and motion compensation are performed on the continuous dynamic images to ensure the spatial consistency and temporal continuity of the multi-frame reconstructed images, so as to output optimized images containing overall details of the organ and enhanced display of suspected lesion areas for analysis of gastrointestinal tract lesions.

7. The method according to claim 2, characterized in that The image acquisition module acquires a low-resolution image sequence of gastrointestinal barium double contrast radiography, including: In response to the trigger instruction of the DR device in the low-dose mode of lowering the continuous pulse X-ray emission power and increasing the dynamic flat panel detection sensitivity, multiple continuous angiographic images are collected in real time; The collected angiographic images are preprocessed to identify and mark insufficiently filled areas of the digestive tract due to physical limitations of the user and insufficient inhalation, and local low-contrast areas due to uneven distribution of barium, to generate the low-resolution image sequence.

8. A super-resolution reconstruction device for digestive tract double contrast imaging of gastrointestinal barium, characterized in that: A control module for a super-resolution reconstruction system for optimizing gastrointestinal double-contrast imaging of gastrointestinal tract as claimed in claim 1, the device comprising: a sequence acquisition unit configured to acquire a low-resolution image sequence acquired by the image acquisition module for gastrointestinal barium double contrast imaging; the low-resolution image sequence being acquired by the DR device under low-dose conditions by lowering the continuous pulse X-ray emission power and increasing the dynamic flat-panel detection sensitivity corresponding to the DR device; the low-resolution image sequence corresponding to imaging under hardware-constrained scenarios, wherein the hardware-constrained scenarios include one or more of the following: an inability to adopt high frame rates, multi-angle imaging, user physical limitations, uneven distribution of barium, and insufficient inhalation by the user; a sequence generation unit configured to dynamically capture long-range dependencies between image elements at each position in the low-resolution image sequence through a self-attention mechanism, thereby generating a feature representation reflecting the contextual association between the sequence images; and generating a high-resolution image sequence containing more detailed information about suspected lesions based on the feature representation and local detail features within a single image in the low-resolution image sequence; The reconstruction completion unit is used to optimize the details of the digestive tract gas-barium double contrast imaging according to the high-resolution image sequence to assist in displaying the microstructure of the digestive tract organ mucosal folds and the cavity morphology, and complete the super-resolution reconstruction of the digestive tract gas-barium double contrast imaging optimization.

9. A control module, characterized in that: The control module includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the method according to any one of claims 2 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program. When the computer-readable instructions are executed by the processor, one or more processors are caused to perform the method according to any one of claims 2 to 7.

Images