A low-dose medical image denoising enhancement system and optimization method

Abstract

The application discloses a low-dose medical image noise reduction enhancement system and an optimization method, relates to the technical field of medical image processing, and comprises the following components: an image input preprocessing module, a noise intelligent typing module, a kernel state acquisition module, a parameter dynamic adjustment module and a noise reduction enhancement feedback module. According to the complete closed-loop processing flow, the application generates strong adaptability of the noise reduction parameter instruction through accurate identification of the mixed noise in the low-dose medical image and real-time sensing of the system state, adopts differential processing to reserve the key details of the image and eliminate noise interference, improves the image definition and diagnostic usability, greatly improves the system adaptability and operation stability, solves the technical problems of the prior art, guarantees efficient and stable output of the system, and reduces resource consumption and processing delay.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a low-dose medical image noise reduction and enhancement system and optimization method. Background Technology

[0002] Low-dose medical imaging technology is widely used in clinical diagnosis. Equipment such as CT and X-rays effectively protect patients' health by reducing radiation dose. However, during low-dose imaging, due to factors such as equipment performance and imaging environment, images are prone to mixed noise such as Gaussian noise and Poisson noise, leading to decreased image clarity and blurred details, affecting medical staff's accurate judgment of lesions. Therefore, developing efficient low-dose medical image denoising and enhancement techniques that suppress noise interference while preserving key diagnostic information has become an important research direction in the field of medical imaging.

[0003] Existing low-dose medical image denoising and enhancement techniques have many shortcomings. Some techniques employ a single denoising algorithm, which cannot accurately identify the type and intensity of mixed noise, resulting in poor denoising effects. While some techniques attempt noise classification, they rely on traditional algorithms, resulting in limited feature extraction capabilities and difficulty in capturing multi-scale noise correlation features. Furthermore, the denoising parameters of existing techniques are mostly fixed settings, failing to consider resource consumption and latency during system operation, easily leading to resource waste or excessive processing delays. In addition, most techniques adopt a globally uniform denoising strategy, failing to distinguish between image edges and non-edge regions, easily losing edge details during denoising and affecting the diagnostic value of the image. Moreover, the lack of an effective feedback adjustment mechanism prevents dynamic parameter optimization based on actual processing results, leading to insufficient system adaptability.

[0004] In summary, existing low-dose medical image denoising and enhancement techniques have significant shortcomings in terms of noise classification accuracy, parameter adjustment flexibility, regional adaptability, and feedback optimization capabilities, making it difficult to meet the dual demands of clinical diagnosis for image quality and system operating efficiency. Therefore, there is an urgent need for a denoising and enhancement scheme that can intelligently identify noise characteristics, dynamically adapt to system states, layer-wise protect image details, and achieve closed-loop optimization. This would improve the diagnostic reliability of low-dose medical images while ensuring stable and efficient system operation, providing strong support for clinical diagnosis. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a low-dose medical image denoising and enhancement system and optimization method. It can output standardized data through preprocessing of the original image; a noise intelligent classification module extracts features and quantifies noise; a kernel state acquisition module acquires the system state; a parameter dynamic adjustment module generates denoising parameters in combination with a rule base; and a denoising and enhancement feedback module performs differentiated denoising and enhancement, outputs high-quality images, and feeds back data to optimize parameters.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, a low-dose medical image noise reduction and enhancement system, the system comprising the following components: an image input preprocessing module, a noise intelligent classification module, a kernel state acquisition module, a parameter dynamic adjustment module, and a noise reduction and enhancement feedback module; The image input preprocessing module is used to receive the raw images output by the low-dose medical imaging device, perform format parsing, data verification, normalization, block zero padding, and preliminary noise filtering on the raw images, and output standardized image data. The noise intelligent classification module is connected to the image input preprocessing module, receives standardized image data, extracts features from the standardized image data through a pre-trained neural network model, identifies noise type and quantizes intensity through a noise classification intensity quantization algorithm, and outputs noise feature data and trigger signal. The kernel status acquisition module: responds to trigger signals, acquires resource usage data and latency data during system operation through trigger signals, processes the acquired data, and outputs system status data; The parameter dynamic adjustment module is connected to the noise intelligent classification module and the kernel status acquisition module, and receives noise feature data and system status data; it uses the kernel feedback noise reduction parameter adaptive algorithm and combines it with a preset association mapping rule library to generate noise reduction parameter instructions. The noise reduction and enhancement feedback module is connected to the parameter dynamic adjustment module, receives noise reduction parameter instructions, performs image edge region recognition, layered noise reduction and image enhancement processing, outputs the noise-reduced and enhanced medical image, and feeds back its own operating data as feedback data to the parameter dynamic adjustment module.

[0007] Furthermore, the process of format parsing, data verification, normalization, block zero-padding, and preliminary noise filtering in the image input preprocessing module to output standardized image data is as follows: The file format of the original image is identified and parsed to extract valid pixel data and imaging metadata; the parsed data undergoes integrity verification to check for missing pixel blocks or metadata anomalies, identifying abnormal pixels; the verified pixel data is normalized, and the normalized pixel data image is divided into fixed-size image sub-blocks according to the requirements of noise identification and denoising; zero pixels are added to image sub-blocks with insufficient size at the edges to ensure the consistency of sub-block specifications; preliminary noise suppression is performed on the segmented image, preserving the basic image structure and edge information, and all processed image sub-blocks are integrated to output standardized image data.

[0008] Furthermore, the pre-trained neural network model in the noise intelligent classification module is a lightweight convolutional neural network, including an input layer, a multi-scale convolutional layer, a batch normalization layer, an activation layer, and a global pooling layer. The process of feature extraction from standardized image data using the pre-trained neural network model is as follows: After the standardized image data is input through the input layer, the multi-scale convolutional layer uses convolutional kernels of different sizes to perform multi-dimensional feature perception on the image to capture noise-related features at different scales; after the noise-related features are calibrated by the batch normalization layer, they are input into the activation layer for feature enhancement; the enhanced features are then compressed in dimension and aggregated by the global pooling layer to complete the feature extraction of the standardized image data.

[0009] Furthermore, the calculation formula for the noise mixing classification intensity quantization algorithm in the noise intelligent classification module is as follows: ,in, This represents the quantized value of the overall noise mixing intensity of an image sub-block. The width of the image sub-block; The height of the image sub-block; This represents the total number of pixels in the sub-block. These are the two-dimensional spatial coordinates of a pixel; The weights are for Gaussian noise classification. The weights for Poisson noise classification; coordinates Gaussian noise intensity at the location; coordinates The Poisson noise intensity at the location is determined. Based on the noise features extracted by the pre-trained neural network, the gray-scale statistical characteristics of local pixels in the standardized image data are analyzed. Combined with the characteristic distribution patterns of different types of noise, the dominant noise type and the proportion of mixed noise in the image are determined. At the same time, the intensity distribution of each type of noise in different image regions is quantified and output, providing accurate noise information support for the parameter dynamic adjustment module to generate noise reduction parameter instructions in a targeted manner.

[0010] Furthermore, the kernel status acquisition module collects resource usage data and latency data during system operation via trigger signals, processes the collected data, and outputs system status data. The specific steps are as follows: After receiving the trigger signal output by the noise intelligent classification module, it collects resource usage data such as CPU utilization, memory usage, and video memory usage at fixed time intervals of 30 seconds, as well as transmission latency data of standardized image data and latency data of data processing. The collected resource usage data and latency data are preprocessed, and the preprocessed resource usage data and latency data are feature-fused to generate system status data that comprehensively reflects the real-time operating load and response efficiency of the system and outputs it.

[0011] Furthermore, the calculation formula for the kernel feedback noise reduction parameter adaptive algorithm in the parameter dynamic adjustment module is as follows: ,in, This is the instruction set for the final generated noise reduction parameters; These are the weighting coefficients for noise feature data; Noise characteristic data; These are the weighting coefficients for the system state data; This is system status data; The noise reduction parameter offset is used as the baseline. The noise feature data output by the noise intelligent classification module is dynamically integrated with the system status data obtained by the kernel status acquisition module. The generated noise reduction parameter instructions can accurately match the noise characteristics of different regions of the image, so that the noise reduction and enhancement processing can maintain the integrity of the image structure and edge information while maintaining the efficient and low-latency operation of the system, and achieve a dynamic balance between noise reduction effect and system performance.

[0012] Furthermore, the preset association mapping rule base in the parameter dynamic adjustment module is a multi-dimensional structured rule set, including a subset of noise feature-parameter adaptation rules, a subset of system state-constraint rules, a subset of weight coefficient matching rules, and a subset of benchmark parameter mapping rules. The noise feature-parameter adaptation rule subset stores the association logic between different noise types, mixing ratios, and regional intensity distributions and the corresponding noise reduction parameter adjustment directions. The system state-constraint rule subset records the constraint relationship between system CPU utilization, memory usage, and processing latency data and parameter adjustment amplitude. The weight coefficient matching rule subset defines the adaptive adjustment logic of noise feature data weight coefficients and system state data weight coefficients as noise characteristics and system operating states change. The benchmark parameter mapping rule subset stores the benchmark noise reduction parameter offset association information adapted to different low-dose medical image modalities.

[0013] Furthermore, the process of image edge region recognition, layered denoising, and image enhancement processing in the denoising and enhancement feedback module is as follows: Based on the received denoising parameter instructions, a preset edge feature template is invoked to perform global feature matching on the standardized image data. Combined with neighborhood difference analysis, edge regions and non-edge regions are located and marked. Layered denoising is performed based on the marking results. A weak denoising strategy is used for edge regions to preserve detail information, while a strong denoising strategy is used for non-edge regions to eliminate noise interference. The denoising process follows the region adaptation requirements in the denoising parameter instructions throughout. Image enhancement processing is performed based on layered denoising. A preset edge enhancement template is used to enhance the contours of the marked edge regions and optimize the region transitions of non-edge regions. After processing, a denoised and enhanced medical image is output. The preset edge feature template is a predefined and stored structured feature set containing various typical edge morphological features in medical images, used for global feature matching of the image to quickly locate and mark edge regions and non-edge regions.

[0014] Furthermore, the process by which the noise reduction and enhancement feedback module feeds its own operational data back to the parameter dynamic adjustment module is as follows: During the entire process of image edge region recognition, layered noise reduction, and image enhancement, the noise reduction and enhancement feedback module collects operational data generated during its operation in real time, including total processing time, completion rate of each stage, compliance data of noise reduction and enhancement operations, and result quality assessment data. The collected operational data is then structured and redundant information is removed to generate a feedback dataset that conforms to the data received by the parameter dynamic adjustment module. This feedback dataset is then transmitted to the parameter dynamic adjustment module in real time through a data interaction interface, along with a data verification identifier to ensure the integrity and accuracy of the feedback data. Upon receiving the feedback dataset, the parameter dynamic adjustment module stores it in a pre-defined association mapping rule base.

[0015] On the other hand, a low-dose medical image denoising and enhancement optimization method is characterized by the following specific steps: S100, Image Input Preprocessing: Receives the raw image output from the low-dose medical imaging device, performs format parsing, data verification, normalization, block zeroing, and preliminary noise filtering, and outputs standardized image data. S200, Noise Intelligent Classification: Receives standardized image data, extracts features through a pre-trained neural network model, identifies noise types and quantifies intensity, and outputs noise feature data and trigger signals; S300, Kernel Status Acquisition: In response to trigger signals, it collects resource usage and latency data during system operation, processes the data, and outputs system status data; S400, Parameter Dynamic Adjustment: Receives noise characteristic data and system status data, and generates noise reduction parameter instructions by combining them with a preset association mapping rule base; S500, Noise Reduction and Enhancement Feedback: Receives noise reduction parameter instructions, performs image edge region recognition, layered noise reduction and image enhancement processing, outputs the noise-reduced and enhanced medical image, and feeds back its own operating data to the parameter dynamic adjustment step.

[0016] Compared with existing technologies, this low-dose medical image denoising and enhancement system and optimization method have the following beneficial effects: I. This invention achieves accurate identification of mixed noise in low-dose medical images and real-time perception of system operating status by setting up a noise intelligent classification module and a kernel status acquisition module. The noise intelligent classification module uses a specific neural network model to extract multi-dimensional features and combines a dedicated algorithm to complete the quantitative analysis of noise type and intensity, making noise feature capture more comprehensive and accurate. The kernel status acquisition module responds to trigger signals and comprehensively collects system resource usage and latency-related data, providing a reliable basis for parameter adjustment. The parameter dynamic adjustment module integrates noise feature data and system status data, and generates highly adaptable noise reduction parameter instructions based on a multi-dimensional rule base, ensuring that parameter settings not only meet noise suppression requirements but also match the system's operating load. Combined with the region identification and layered noise reduction strategy of the noise reduction enhancement feedback module, different processing methods are used for different regions, effectively preserving key image details while maximizing the elimination of noise interference, significantly improving image clarity and diagnostic usability.

[0017] Second, this invention constructs a complete processing flow of preprocessing-classification-adjustment-feedback, forming a closed-loop optimization mechanism, which greatly improves the adaptability and operational stability of the system. The image input preprocessing module performs a series of standardized processes on the original image, such as format parsing, data verification, and zero padding, providing a high-quality data foundation for subsequent noise identification and noise reduction enhancement, and reducing the impact of data anomalies on the processing effect. The noise reduction enhancement feedback module collects its own operating data in real time while completing image processing, and feeds it back to the parameter dynamic adjustment module after processing, providing data support for the dynamic updating of the rule base, enabling the system to continuously optimize parameter settings according to the actual processing effect. The collaborative work of each module not only solves the problems of insufficient targeting of noise processing and rigid parameter adjustment in traditional technologies, but also ensures the efficient and stable output of the system under different operating scenarios, while reducing the system's operating resource consumption and processing latency.

[0018] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0020] Figure 1 This is a module connection diagram of a low-dose medical image denoising and enhancement system. Figure 2 This is a schematic diagram of collaborative data transmission for noise classification, kernel state acquisition, and parameter adjustment in a low-dose medical image denoising and enhancement system. Figure 3 This is a flowchart of a low-dose medical image denoising and enhancement optimization method. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0022] Example 1 like Figure 1 As shown, the system receives raw DICOM format images from a low-dose CT chest scan device. It first identifies and parses the file format, extracting valid pixel data and CT imaging-related metadata. Integrity checks are performed to identify missing pixels or metadata anomalies in the lung region, accurately identifying abnormal pixels and preventing interference from invalid data in subsequent processing. The verified pixel data is normalized to ensure uniform pixel size for easier subsequent operations. Based on noise identification and denoising requirements, the normalized pixel data image is divided into fixed-size image sub-blocks. Zero pixels are added to image sub-blocks at the chest edges to ensure consistent sub-block dimensions and uniform processing. Preliminary noise suppression is applied to the segmented images, focusing on preserving basic structures and edge information such as pulmonary vessels and bronchi. Finally, all processed image sub-blocks are integrated to output standardized image data, providing a high-quality input foundation for subsequent modules.

[0023] The system receives standardized image data from the image input preprocessing module and extracts features using a pre-trained lightweight convolutional neural network. After the standardized image data is input through the input layer, a multi-scale convolutional layer uses kernels of different sizes to perceive multi-dimensional features, comprehensively capturing noise-related features at different scales. These noise-related features are then processed by a batch normalization layer for data distribution calibration, making the feature representation more stable. An input activation layer enhances the features, and a global pooling layer further compresses and aggregates the enhanced features to improve feature utilization. Based on the aggregated features, a noise grading and intensity quantization algorithm is used to identify the noise type, determining that the CT chest image mainly contains a mixture of Gaussian and Poisson noise. Simultaneously, the intensity of the mixed noise is quantized using the following formula: ,in, This represents the quantized value of the overall noise mixing intensity of an image sub-block. The width of the image sub-block; The height of the image sub-block; This represents the total number of pixels in the sub-block. These are the two-dimensional spatial coordinates of a pixel; The weights are for Gaussian noise classification. The weights for Poisson noise classification; coordinates Gaussian noise intensity at the location; coordinates The Poisson noise intensity at the location; the final output includes noise characteristic data such as noise type, mixing ratio, and regional intensity distribution, providing an accurate basis for parameter adjustment, as well as trigger signals for triggering subsequent modules to ensure the smooth progress of the system process.

[0024] Upon receiving the trigger signal from the noise intelligent classification module, the system continuously collects resource usage and latency data at fixed time intervals. Resource usage data includes CPU utilization, memory utilization, and video memory utilization. Latency data includes the transmission latency of standardized image data from the preprocessing module to this module, as well as the latency of each data processing step, providing a comprehensive understanding of the system's real-time operating status. The collected resource usage and latency data are preprocessed to remove redundant interference. The two types of data are then fused to generate and output system status data that comprehensively reflects the current system's real-time operating load and response efficiency, allowing parameter adjustment to fully adapt to the current system's operating capabilities.

[0025] The system receives noise feature data from the noise intelligent classification module and system status data from the kernel status acquisition module. It then calls the kernel feedback noise reduction parameter adaptive algorithm and, combined with a preset association mapping rule library, generates noise reduction parameter instructions using the following formula: ,in, This is the instruction set for the final generated noise reduction parameters; These are the weighting coefficients for noise feature data; Noise characteristic data; These are the weighting coefficients for the system state data; This is system status data; This serves as the baseline denoising parameter offset. A subset of noise feature-parameter adaptation rules from the pre-defined association mapping rule base determines the corresponding denoising parameter adjustment direction based on the noise type, mixing ratio, and regional intensity distribution of the current CT chest image, ensuring the parameters closely match the noise characteristics. A subset of system state-constraint rules limits the parameter adjustment range based on collected CPU usage, memory occupancy, and processing latency data to avoid exceeding the system's capacity. A subset of weight coefficient matching rules adaptively adjusts the weight coefficients of noise feature data and system state data based on the current noise characteristics and system operating status, balancing their influence. A subset of baseline parameter mapping rules provides a baseline denoising parameter offset adapted to the CT image modality. Finally, these are integrated to generate targeted denoising parameter instructions, providing accurate guidance for subsequent operations.

[0026] The system receives noise reduction parameter instructions from the dynamic parameter adjustment module, calls a preset edge feature template to perform global feature matching on the standardized image data, and combines neighborhood difference analysis to accurately locate and mark edge regions such as pulmonary blood vessel edges and bronchial edges, as well as non-edge regions such as lung parenchyma, ensuring more targeted layered noise reduction. Layered noise reduction is performed based on the marking results. A weak noise reduction strategy is used for edge regions to preserve details to the maximum extent, while a strong noise reduction strategy is used for non-edge regions to effectively eliminate noise interference. The entire noise reduction process follows the region adaptation requirements in the noise reduction parameter instructions. Image enhancement processing is then performed on the basis of layered noise reduction. A preset edge enhancement template is used to enhance the contours of the marked edge regions to make lesions easier to identify, and region transition optimization is performed on non-edge regions to improve the overall image coordination. After processing, a noise-reduced and enhanced CT chest image is output to meet clinical diagnostic needs. Simultaneously, the system collects its own operational data in real time throughout the entire process, including total processing time, completion rate of each stage, compliance data for noise reduction and enhancement operations, and result quality assessment data. The collected operational data is structured and redundant information is removed to generate a feedback dataset that meets the data reception requirements of the parameter dynamic adjustment module. The feedback dataset is transmitted to the parameter dynamic adjustment module in real time through the data interaction interface, with a data verification mark attached to ensure the integrity and accuracy of the feedback data. After receiving the feedback dataset, the parameter dynamic adjustment module stores it in association with a preset association mapping rule base, providing a better reference for subsequent similar image processing.

[0027] Example 2 like Figure 3 As shown, the system receives raw DICOM format images from a low-dose X-ray knee joint bone scanning device. First, it identifies and parses the file format, extracting valid pixel data and X-ray imaging metadata of the knee joint bones and surrounding soft tissues. Integrity checks are performed to identify missing pixel blocks or metadata anomalies in the knee joint bone region, accurately identifying abnormal pixels and reducing interference from invalid data in subsequent processing. The verified pixel data is normalized to unify data scale for easier collaborative processing. Based on the requirements of noise identification and denoising, the normalized pixel data image is divided into fixed-size image sub-blocks. Zero pixels are added to image sub-blocks with insufficient size at the knee joint edges and bone gaps to ensure uniform sub-block size and consistent processing. Preliminary noise suppression is performed on the segmented images, focusing on preserving the knee joint bone contour, epiphyseal lines, and other basic structural and edge information. Then, all processed image sub-blocks are integrated, outputting standardized image data to provide reliable support for subsequent noise classification and parameter adjustment.

[0028] The system receives standardized image data from the image input preprocessing module and extracts features using a pre-trained lightweight convolutional neural network. After the standardized image data is input into the input layer, a multi-scale convolutional layer uses kernels of different sizes to perceive multi-dimensional features of the image, comprehensively capturing noise-related features at different scales. These noise-related features are then calibrated through a batch normalization layer to make the feature representation more stable. An input activation layer enhances the features, and a global pooling layer further compresses and aggregates the enhanced features to improve feature utilization. Based on the aggregated features, a noise-mixing classification intensity quantization algorithm is used to identify the noise type. It is determined that the X-ray knee joint image is mainly composed of Poisson noise mixed with a small amount of Gaussian noise. Simultaneously, the intensity of the mixed noise is quantized, outputting noise feature data including noise type, mixing ratio, and regional intensity distribution. This provides accurate data for parameter adjustment and triggers timely linkage with the kernel state acquisition module to ensure the orderly connection of the system process.

[0029] Upon receiving the trigger signal from the noise intelligent classification module, the system collects resource usage and latency data at fixed time intervals. Resource usage data includes CPU utilization, memory utilization, and video memory utilization. Latency data includes transmission latency of standardized image data and latency data from each data processing stage, providing a comprehensive understanding of the system's real-time operation. The collected resource usage and latency data are preprocessed to remove invalid data. The two types of data are then fused to generate and output system status data that comprehensively reflects the system's real-time operating load and response efficiency, allowing parameter adjustment to optimize settings based on the system's actual capabilities.

[0030] The system receives noise feature data and system status data, invokes the kernel feedback denoising parameter adaptive algorithm, and generates denoising parameter instructions based on a preset association mapping rule library. The noise feature-parameter adaptation rule subset in the preset association mapping rule library, based on the noise type, mixing ratio, and regional intensity distribution of the current X-ray knee joint image, clarifies the direction of denoising parameter adjustment to align with the noise characteristics of the skeletal image. The system status-constraint rule subset, based on CPU usage, memory occupancy, and processing latency data, limits the parameter adjustment range to adapt to the system's operating capabilities. The weight coefficient matching rule subset adaptively adjusts the weight coefficients of noise feature data and system status data to balance their influence based on the current noise characteristics and system operating status. The baseline parameter mapping rule subset provides a baseline denoising parameter offset adapted to the X-ray skeletal image modality. Finally, it generates denoising parameter instructions adapted to this scenario, providing scientific guidance for subsequent operations, such as... Figure 2 As shown.

[0031] The system receives denoising parameter instructions from the dynamic parameter adjustment module, calls a preset edge feature template to perform global feature matching on the standardized image data, and combines neighborhood difference analysis to accurately locate and mark edge regions such as the knee joint bone contour edge and epiphyseal line edge, as well as non-edge regions such as surrounding soft tissue, ensuring more targeted layered denoising. Layered denoising is performed based on the marking results, using a weak denoising strategy for edge regions to preserve bone details and a strong denoising strategy for non-edge regions to eliminate noise interference. The denoising process strictly follows the region adaptation requirements in the denoising parameter instructions. Image enhancement processing is then performed on the basis of layered denoising. A preset edge enhancement template is used to enhance the contours of the marked edge regions to make the bone structure clearer, and region transition optimization is performed on non-edge regions to improve the overall image quality. After processing, a denoised and enhanced X-ray medical image of the knee joint is output to meet clinical diagnostic needs. Simultaneously, the system collects its own operational data in real time throughout the entire process, including total processing time, completion rate of each stage, compliance data for noise reduction and enhancement operations, and result quality assessment data. The collected operational data is structured and redundant information is removed to generate a feedback dataset that meets the data receiving requirements of the parameter dynamic adjustment module. The feedback dataset is transmitted to the parameter dynamic adjustment module in real time through the data interaction interface, with a data verification mark attached to ensure the integrity and accuracy of the feedback data. After receiving the feedback dataset, the parameter dynamic adjustment module associates and stores it in a preset association mapping rule base, continuously enriching the rule base content to help improve system performance.

[0032] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A low-dose medical image noise reduction and enhancement system, characterized in that, The system comprises the following components: an image input preprocessing module, a noise intelligent classification module, a kernel status acquisition module, a parameter dynamic adjustment module, and a noise reduction and enhancement feedback module; The image input preprocessing module is used to receive the raw images output by the low-dose medical imaging device, perform format parsing, data verification, normalization, block zero padding, and preliminary noise filtering on the raw images, and output standardized image data. The noise intelligent classification module is connected to the image input preprocessing module, receives standardized image data, extracts features from the standardized image data through a pre-trained neural network model, identifies noise type and quantizes intensity through a noise classification intensity quantization algorithm, and outputs noise feature data and trigger signal. The kernel status acquisition module: responds to trigger signals, acquires resource usage data and latency data during system operation through trigger signals, processes the acquired data, and outputs system status data; The parameter dynamic adjustment module is connected to the noise intelligent classification module and the kernel status acquisition module, and receives noise feature data and system status data; it uses the kernel feedback noise reduction parameter adaptive algorithm and combines it with a preset association mapping rule library to generate noise reduction parameter instructions. The noise reduction and enhancement feedback module is connected to the parameter dynamic adjustment module, receives noise reduction parameter instructions, performs image edge region recognition, layered noise reduction and image enhancement processing, outputs the noise-reduced and enhanced medical image, and feeds back its own operating data as feedback data to the parameter dynamic adjustment module.

2. The low-dose medical image denoising and enhancement system according to claim 1, characterized in that, The process of the image input preprocessing module performing format parsing, data verification, normalization, block zero padding, and preliminary noise filtering on the original image and outputting standardized image data is as follows: identifying and parsing the file format of the original image, and extracting effective pixel data and imaging metadata. The parsed data is subjected to integrity verification to check for missing pixel blocks or metadata anomalies and to identify abnormal pixels. The verified pixel data is normalized and the normalized pixel data image is divided into fixed-size image sub-blocks according to the requirements of noise recognition and noise reduction. Zero pixels are added to image sub-blocks with insufficient size at the edges to ensure the consistency of sub-block specifications. Preliminary noise suppression is performed on the segmented image to preserve the basic structure and edge information of the image, and all processed image sub-blocks are integrated to output standardized image data.

3. The low-dose medical image denoising and enhancement system according to claim 1, characterized in that, The pre-trained neural network model in the noise intelligent classification module is a lightweight convolutional neural network, including an input layer, a multi-scale convolutional layer, a batch normalization layer, an activation layer, and a global pooling layer. The process of feature extraction of standardized image data by the pre-trained neural network model is as follows: after the standardized image data is input through the input layer, the multi-scale convolutional layer uses convolutional kernels of different sizes to perform multi-dimensional feature perception on the image in order to capture noise-related features at different scales. After the data distribution of the noisy associated features is calibrated by the batch normalization layer, they are input into the activation layer for feature enhancement; the enhanced features are then subjected to dimensionality compression and information aggregation through the global pooling layer.

4. The low-dose medical image denoising and enhancement system according to claim 1, characterized in that, The calculation formula for the noise mixing classification intensity quantization algorithm in the noise intelligent classification module is as follows: ,in, This represents the quantized value of the overall noise mixing intensity of an image sub-block. The width of the image sub-block; The height of the image sub-block; This represents the total number of pixels in the sub-block. These are the two-dimensional spatial coordinates of a pixel; The weights are for Gaussian noise classification. The weights for Poisson noise classification; coordinates Gaussian noise intensity at the location; coordinates The Poisson noise intensity at that location.

5. The low-dose medical image denoising and enhancement system according to claim 1, characterized in that, The kernel status acquisition module collects resource usage data and latency data during system operation through trigger signals, processes the collected data, and outputs system status data. The specific steps are as follows: after receiving the trigger signal output by the noise intelligent classification module, the module collects resource usage data such as CPU utilization, memory usage, and video memory usage at fixed time intervals, as well as transmission latency data of standardized image data and latency data of data processing. The collected resource usage data and latency data are preprocessed, and the preprocessed resource usage data and latency data are feature fused to generate system status data that comprehensively reflects the real-time operating load and response efficiency of the system and output it.

6. The low-dose medical image denoising and enhancement system according to claim 1, characterized in that, The calculation formula for the kernel feedback noise reduction parameter adaptive algorithm in the parameter dynamic adjustment module is as follows: ,in, This is the instruction set for the final generated noise reduction parameters; These are the weighting coefficients for noise feature data; Noise characteristic data; These are the weighting coefficients for the system state data; This is system status data; This is the offset of the baseline noise reduction parameters.

7. The low-dose medical image denoising and enhancement system according to claim 1, characterized in that, The preset association mapping rule base in the parameter dynamic adjustment module is a multi-dimensional structured rule set, including a subset of noise feature-parameter adaptation rules, a subset of system state-constraint rules, a subset of weight coefficient matching rules, and a subset of benchmark parameter mapping rules; The noise characteristics - parameter adaptation rule subset stores the correlation logic between different noise types, mixing ratios, and regional intensity distributions and the corresponding noise reduction parameter adjustment directions; the system status - constraint rule subset records the constraint relationship between system CPU utilization, memory usage, and processing latency data and parameter adjustment ranges. The weight coefficient matching rule subset defines the adaptive adjustment logic of the weight coefficients of noise feature data and system state data as noise characteristics and system operating state change; the benchmark parameter mapping rule subset stores the benchmark denoising parameter offset association information adapted to different low-dose medical image modalities.

8. The low-dose medical image denoising and enhancement system according to claim 1, characterized in that, The noise reduction and enhancement feedback module performs image edge region recognition, layered noise reduction, and image enhancement as follows: Based on the received noise reduction parameter instructions, a preset edge feature template is invoked to perform global feature matching on the standardized image data. Combined with neighborhood difference analysis, edge regions and non-edge regions are located and marked. Layered noise reduction is performed based on the marking results. A weak noise reduction strategy is used for edge regions to preserve detail information, while a strong noise reduction strategy is used for non-edge regions to eliminate noise interference. The entire noise reduction process follows the region adaptation requirements in the noise reduction parameter instructions. Image enhancement is then performed based on the layered noise reduction. A preset edge enhancement template is used to enhance the contours of the marked edge regions and optimize the region transition of non-edge regions. After processing, the noise-reduced and enhanced medical image is output.

9. A low-dose medical image denoising and enhancement system according to claim 1, characterized in that, The process by which the noise reduction and enhancement feedback module feeds its own operational data back to the parameter dynamic adjustment module is as follows: During the entire process of image edge region recognition, layered noise reduction, and image enhancement, the noise reduction and enhancement feedback module collects operational data generated during its operation in real time, including total processing time, completion rate of each stage, compliance data of noise reduction and enhancement operations, and result quality evaluation data. The collected operational data is then structured and redundant information is removed to generate a feedback dataset that meets the requirements of the parameter dynamic adjustment module. This feedback dataset is then transmitted to the parameter dynamic adjustment module in real time via a data interaction interface, along with a data verification identifier to ensure the integrity and accuracy of the feedback data. Upon receiving the feedback dataset, the parameter dynamic adjustment module stores it in a pre-defined association mapping rule base.

10. A low-dose medical image denoising and enhancement optimization method, applicable to the low-dose medical image denoising and enhancement system according to any one of claims 1-9, characterized in that, The specific steps of this method are as follows: S100, Image Input Preprocessing: Receives the raw image output from the low-dose medical imaging device, performs format parsing, data verification, normalization, block zeroing, and preliminary noise filtering, and outputs standardized image data. S200, Noise Intelligent Classification: Receives standardized image data, extracts features through a pre-trained neural network model, identifies noise types and quantifies intensity, and outputs noise feature data and trigger signals; S300, Kernel Status Acquisition: In response to trigger signals, it collects resource usage and latency data during system operation, processes the data, and outputs system status data; S400, Parameter Dynamic Adjustment: Receives noise characteristic data and system status data, and generates noise reduction parameter instructions by combining them with a preset association mapping rule base; S500, Noise Reduction and Enhancement Feedback: Receives noise reduction parameter instructions, performs image edge region recognition, layered noise reduction and image enhancement processing, outputs the noise-reduced and enhanced medical image, and feeds back its own operating data to the parameter dynamic adjustment step.

Images