Medical image denoising system and method based on 3D chaos and attraction-repulsion algorithm

Abstract

This invention relates to the field of image processing technology, specifically to a medical image denoising system and method based on 3D chaos and attraction-repulsion algorithms. The method includes determining the optimal parameter set for each filter corresponding to the hybrid filter used for denoising the medical image based on historical raw medical images and a preset hybrid filter parameter set optimization strategy, and configuring the optimal parameter set into the corresponding filters. The preset hybrid filter parameter set optimization strategy includes a 3D-LY chaotic mapping strategy for initializing the population corresponding to the hybrid filter parameter set and an improved attraction-repulsion optimization strategy for optimizing the initialized population. The method involves acquiring the current medical image to be processed at the current moment, adding noise to the acquired current medical image to generate a corresponding noisy current image, and using the noisy current image as input data into the hybrid filter to output the corresponding denoised medical image.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a medical image denoising system and method based on 3D chaos and attraction-repulsion algorithms. Background Technology

[0002] Medical imaging plays a crucial role in disease diagnosis, but it is susceptible to interference from factors such as motion artifacts and electronic noise from equipment, resulting in speckle noise, salt-and-pepper noise, and other forms of noise. Noise not only degrades image visual quality but can also mask lesion features, leading to misdiagnosis or missed diagnosis, and even affecting the accuracy of subsequent automated analyses (such as lesion segmentation and volume measurement). Therefore, efficient denoising techniques are one of the core requirements for medical image processing.

[0003] Existing technologies typically employ a single filter for denoising or use optimization algorithms. However, single-filter denoising has inherent drawbacks. For example, while linear filters can effectively suppress Gaussian noise, they can lead to edge blurring and texture loss, making it difficult to handle complex noise patterns and resulting in poor denoising performance.

[0004] Existing optimization algorithms have the following problems:

[0005] Traditional random initialization leads to uneven parameter space coverage and a tendency to get trapped in local optima. For example, one-dimensional chaotic maps (such as logistic maps) have insufficient fractal dimension, resulting in clustered population distribution. It lacks the ability to dynamically balance global exploration and local exploitation, making it difficult to dynamically balance noise suppression and detail preservation, thus leading to poor denoising results.

[0006] Therefore, there is an urgent need for a medical image denoising system and method based on 3D chaos and attraction-repulsion algorithms, which can solve the problem of poor denoising effect in existing medical images and greatly improve the effect and efficiency of medical image denoising. Summary of the Invention

[0007] One of the objectives of this invention is to provide a medical image denoising system and method based on 3D chaos and attraction-repulsion algorithms, which can solve the problem of poor denoising effect in medical images in the prior art and greatly improve the effect and efficiency of medical image denoising.

[0008] To achieve the above objectives, a medical image denoising method based on 3D chaos and attraction-repulsion algorithms is provided, comprising the following steps:

[0009] S1. Based on historical original medical images and a preset hybrid filter parameter group optimization strategy, determine the optimal parameter group for each filter corresponding to the hybrid filter used for denoising the medical images, and configure the optimal parameter group into the corresponding filter; the preset hybrid filter parameter group optimization strategy includes a 3D-LY chaotic mapping strategy for initializing the population corresponding to the hybrid filter parameter group and an improved attraction-repulsion optimization strategy for optimizing the initialized population.

[0010] S2. Obtain the current medical image to be processed at the current moment, and add noise to the obtained current medical image to generate the corresponding current noisy image;

[0011] S3. Take the current noisy image as input data, input it into the mixing filter, and output the corresponding denoised medical image.

[0012] The technical principles and effects of this scheme are as follows: Firstly, using historical raw medical images and a hybrid filter parameter set optimization strategy, the optimal candidate groups for each filter in the hybrid filter are determined. The corresponding hybrid filter parameter set optimization strategy includes a 3D-LY chaotic mapping strategy for initializing the population corresponding to the hybrid filter parameter set, and an improved attraction-repulsion optimization strategy for optimizing the initialized population. The 3D-LY chaotic mapping strategy initializes the population corresponding to the parameter set. The chaotic system is highly sensitive to initial values ​​and parameters, generating a non-periodic, ergodic chaotic sequence in three-dimensional space, avoiding the uneven population distribution problem caused by traditional random initialization. Through the randomness and ergodicity of the chaotic sequence, the initial population is evenly distributed in the parameter search space, covering more potential optimal solution regions and providing high-quality initial samples for subsequent optimization.

[0013] Then, by using an improved attraction-repulsion optimization strategy, the attraction (cooperation) and repulsion (competition) behaviors between individuals in a biological population are simulated, thereby adjusting the position of individuals in the population and adjusting their own parameters, thus improving search efficiency and global optimization capabilities.

[0014] The system then acquires the current medical image to be processed at the current moment and adds noise to it, generating the corresponding noisy image. Artificially added simulated noise (such as Gaussian noise or salt-and-pepper noise) simulates the noise environment of real medical images (such as imaging equipment noise or signal transmission interference), making the noise components in the image more obvious and easier to identify.

[0015] Finally, the noisy image is input into the optimized hybrid filter. Each component processes the image in a preset order or in parallel. Through the synergistic effect of the parameters, multi-level noise filtering is achieved, and the denoised image is finally output.

[0016] Compared to existing technologies that rely on manual parameter tuning and single optimization algorithms to adjust filter parameters, this solution employs a hybrid optimization approach combining a 3D-LY chaotic system with an improved attraction-repulsion algorithm. This significantly improves the efficiency and reliability of global optimization, resulting in substantial enhancements in noise suppression, detail preservation, and generalization capabilities for medical image denoising. Its core value lies in combining bio-inspired optimization with image processing techniques, providing an adaptive and efficient solution for complex medical noise environments. It possesses strong clinical application potential and algorithmic scalability, effectively addressing the problem of poor denoising performance in existing medical image denoising technologies and greatly improving the effectiveness and efficiency of medical image denoising.

[0017] Furthermore, the 3D-LY chaotic mapping strategy is as follows:

[0018] Construct a corresponding three-dimensional chaotic model based on the number of filter types in the hybrid filter;

[0019] The three-dimensional chaos model is as follows:

[0020]

[0021] In the formula, , , These are the system state variables in the corresponding three-dimensional chaotic model. , These are the parameters of the three-dimensional chaotic model;

[0022] Based on the system state variables in the three-dimensional chaotic model, the filter parameters corresponding to each filter in the hybrid filter are mapped to one of the system state variables respectively;

[0023] The three-dimensional chaotic model is run and iterated multiple times, and the system state variables corresponding to each iteration are recorded to form a corresponding chaotic sequence matrix; each row of the chaotic sequence matrix corresponds to a filter parameter group.

[0024] Normalize each row of the chaotic sequence in the resulting chaotic sequence matrix to form the individuals of the corresponding initial population, and merge the individuals to form the corresponding initial population.

[0025] Beneficial effects: In this scheme, for the multi-parameter optimization requirements of hybrid filters (such as n filters corresponding to n parameter groups, each group containing m parameters), a three-dimensional chaotic system (x, y, z) is constructed. A complex chaotic sequence is generated through the nonlinear coupling of three state variables. Each variable can be mapped to different types of filter parameters (such as x corresponding to the median filter window size, y corresponding to the Gaussian filter standard deviation, and z corresponding to the wavelet threshold). The cosine / sine function introduces periodicity and nonlinearity, making the system state variables highly sensitive to the initial values ​​and parameters (a, b). The cross term enhances the trajectory complexity in three-dimensional space and avoids falling into simple periodic oscillations.

[0026] Compared to traditional random initialization, where the randomly generated parameter sets may be concentrated in local regions, leading to local optima, this scheme uses a three-dimensional chaotic trajectory to almost uniformly traverse the parameter space, ensuring that the initial population covers various parameter combinations from low to high, providing rich samples for global optimization.

[0027] Meanwhile, based on the requirement that the parameters of different filters need to work together, parameter sets with implicit parameter correlations can be generated through the cross-influence of x, y, and z in the three-dimensional chaotic model. Based on the coupling characteristics of the chaotic model, the potential cooperative rules among the parameters can be automatically mined. Moreover, the generation of chaotic sequences only requires multiple iterations of the three-dimensional chaotic model, which greatly improves the efficiency of generating high-quality initial solutions.

[0028] Furthermore, the improved attraction-repulsion optimization strategy is as follows:

[0029] Step 1: Obtain historical raw medical images and add noise to the raw medical images to generate corresponding raw noisy medical images;

[0030] Step 2: Based on the formed population, configure the filter parameter groups corresponding to the individuals in the population into the hybrid filter;

[0031] Step 3: The original noisy medical image is used as input data and input into each of the hybrid filters configured with the corresponding filter parameter groups to denoise the original noisy medical image and output the original denoised medical image corresponding to each individual in the population.

[0032] Step 4: Calculate the fitness of parameters for each individual population based on the original denoised medical images corresponding to each individual population.

[0033] Step 5: Based on the parameter fitness of each individual in the population, sort them in descending order of parameter fitness. Individuals with the highest parameter fitness in the first preset percentage are selected as global optimal solution candidates, and an attraction mechanism is triggered to drive other individuals in the population to move towards the optimal solution candidate. Individuals with the lowest parameter fitness in the second preset percentage are selected as inferior solution candidates, and a repulsion mechanism is triggered to drive other individuals in the population away from the inferior solution candidate.

[0034] Step 6: Based on the current iteration number, determine the local search strategy corresponding to the iteration number, and search for each individual in the population based on the determined local search strategy to form new individuals after optimization of each individual in the population. Then, repeat step 2 until the iteration requirements are met, and output the current global optimal solution as the optimal parameter set of the filter corresponding to the hybrid filter.

[0035] Beneficial effects: In this scheme, the global optimal solution candidate and inferior solution candidate corresponding to each iteration number are determined by calculating the parameter fitness. Then, the individuals in the population are driven to diffuse and move through the bidirectional pressure corresponding to the attraction and repulsion mechanisms. The attraction promotes the population to gather in the current optimal region, while the repulsion prevents the population from gathering in inferior regions, forming a dynamic balance of "seeking the best and avoiding the worst". That is, it ensures that the characteristics of the optimal solution in each generation can be quickly diffused into the entire population, avoiding the dilution of high-quality solutions due to random mutation. At the same time, by identifying inferior solutions, other individuals are forced to stay away, avoiding the pollution of the population by low-quality parameter groups.

[0036] Of course, the search mode is dynamically switched according to the iteration progress. In the early stage of optimization, the exploration range is expanded, and in the later stage, the focus is on the neighborhood of the optimal solution. This achieves efficient search with "breadth first and depth later", which further optimizes the parameter set and improves the reliability and accuracy of finding the filter parameter set.

[0037] Furthermore, the fitness parameters corresponding to each individual in the population are calculated as follows:

[0038]

[0039]

[0040]

[0041]

[0042] In the formula, For the fitness parameters of an individual in the population, These are the weighting coefficients. This represents the mean square error for each individual. Peak signal-to-noise ratio for an individual. It is a structural similarity index between individuals; For noise-free raw medical images in coordinates Pixel value at that location, To denoise the medical image in coordinates Pixel value at that location, For the number of rows and columns of the image, The maximum grayscale value of the image. Original medical images, For denoising medical images, The pixel mean of the original medical image. The pixel mean of the denoised medical image. The pixel variance of the original medical image. To reduce the pixel variance of denoised medical images, The covariance between the original medical image and the denoised medical image. , is the stability constant.

[0043] Beneficial effects: Compared to existing single-metric evaluations, this scheme, through the negative feedback mechanisms of PSNR and SSIM, forces the parameter set to find a balance between denoising intensity and structure preservation. The gradient of the fitness function with respect to the parameters can be decomposed into a weighted sum of the gradients of each metric. This gradient guides the attraction-repulsion algorithm to search in the direction of simultaneous improvement of multiple metrics, avoiding local optima of a single metric.

[0044] Furthermore, the attraction mechanism in step five is as follows:

[0045] Each individual in the population Subject to global optimal solution candidates Driven by attraction, the corresponding update formula is:

[0046]

[0047] In the formula, Let be the attraction strength constant. This is a dynamic equilibrium factor that decreases linearly with the iteration number t until it reaches 0.1. A random vector between 0 and 1 A random number between 0 and 1. To switch the attraction threshold, The attractive displacement corresponding to the attractive force.

[0048] Beneficial effects: In this scheme, the movement of individuals towards the global optimum is achieved by using global optimum candidates, thus realizing adaptive adjustment of individuals towards the global optimum. Random points in the neighborhood of the optimal solution are generated, which allows individuals to maintain the exploratory nature of their movement when moving towards the global optimal solution. The dynamic balance factor avoids abrupt changes in the strength of attraction, allowing the population to smoothly transition from rapid aggregation to local exploration, and preventing the population from spreading out of control due to a sudden drop in attraction.

[0049] Furthermore, the local search strategies in step six include Brownian motion strategies, trigonometric function strategies, and random jump strategies.

[0050] Beneficial effects: This scheme achieves adaptive search throughout the entire cycle of parameter group optimization by setting multiple local search strategies, thereby achieving high reliability of local search.

[0051] The present invention also provides a medical image denoising system based on 3D chaos and attraction-repulsion algorithm, using the above-mentioned medical image denoising method based on 3D chaos and attraction-repulsion algorithm. Attached Figure Description

[0052] Figure 1 This is a flowchart of the medical image denoising method based on 3D chaos and attraction-repulsion algorithm in Embodiment 1 of the present invention;

[0053] Figure 2 This is the Lyapunov index spectrum under the variation of parameter b in Embodiment 1 of the present invention.

[0054] Figure 3 This is a bifurcation diagram showing the variation of parameter b in Embodiment 1 of the present invention.

[0055] Figure 4 This is the phase diagram of the chaotic mapping in Embodiment 1 of the present invention.

[0056] Figure 5 This is a diagram illustrating the Shannon entropy projected onto the phase space of a chaotic system in Embodiment 1 of the present invention.

[0057] Figure 6 This is a diagram illustrating the Shannon entropy of the phase space projection of the 3D-LY chaotic system in Embodiment 1 of the present invention.

[0058] Figure 7 This is a comparison experiment of noise reduction under speckle noise in Embodiment 1 of the present invention and a certain prior art.

[0059] Figure 8 This is a comparison experiment of noise reduction under salt-and-pepper noise in Embodiment 1 of the present invention and a certain prior art.

[0060] Figure 9 This is a comparison experiment of noise reduction under speckle noise in Embodiment 1 of the present invention and some existing technologies.

[0061] Figure 10This is a comparison experiment of noise reduction under salt-and-pepper noise in Embodiment 1 of the present invention and some existing technologies. Detailed Implementation

[0062] The following detailed description illustrates the specific implementation method:

[0063] Example 1

[0064] Medical image denoising methods based on 3D chaos and attraction-repulsion algorithms are basically as follows: Figure 1 As shown, it includes the following steps:

[0065] S1. Based on historical original medical images and a preset hybrid filter parameter group optimization strategy, determine the optimal parameter group for each filter corresponding to the hybrid filter used for denoising the medical images, and configure the optimal parameter group into the corresponding filter; the preset hybrid filter parameter group optimization strategy includes a 3D-LY chaotic mapping strategy for initializing the population corresponding to the hybrid filter parameter group and an improved attraction-repulsion optimization strategy for optimizing the initialized population.

[0066] The 3D-LY chaotic mapping strategy is as follows:

[0067] Construct a corresponding three-dimensional chaotic model based on the number of filter types in the hybrid filter;

[0068] The three-dimensional chaos model is as follows:

[0069]

[0070] In the formula, , , These are the system state variables in the corresponding three-dimensional chaotic model. , The parameters are those of the three-dimensional chaotic model. In this embodiment, the initial conditions of the system are set to (0, 0.0001, 0.001), and the system parameter a = 7.72 is fixed. The value of parameter b is changed. When parameter b changes at (1, 6), the Lyapunov exponent spectrum of the system with respect to parameter b and its bifurcation diagram are as follows. Figure 2 and Figure 3 As shown. From Figure 2 and Figure 3 It can be seen that when the parameter b is in the range of [1,6], the system is always in a chaotic state.

[0071] certainly Figure 4 For system parameters a = 7.72 and b = 4, determine the phase diagram corresponding to the initial system value (0, 0.0001, 0.001). Figure 4 -a is 3d, Figure 4 -b represents the xy plane. Figure 4 -c represents the xz plane. Figure 4 -d represents the yz plane.

[0072] Based on the system state variables in the three-dimensional chaotic model, the filter parameters corresponding to each filter in the hybrid filter are mapped to one of the system state variables respectively;

[0073] The three-dimensional chaotic model is run and iterated multiple times, and the system state variables corresponding to each iteration are recorded to form a corresponding chaotic sequence matrix; each row of the chaotic sequence matrix corresponds to a filter parameter group.

[0074] Normalize each row of the chaotic sequence in the resulting chaotic sequence matrix to form the individuals of the corresponding initial population, and merge the individuals to form the corresponding initial population.

[0075] The improved attraction-repulsion optimization strategy is as follows:

[0076] Step one involves acquiring historical raw medical images and adding noise to them to generate corresponding raw noisy medical images. In this embodiment, the data used to implement the proposed algorithm and analyze the image preprocessing performance is collected from publicly available brain MRI images on Kaggle. The dataset contains approximately 15,000 2D brain MRI slices, with a size of 496×248, a bit depth of 24, a horizontal resolution of 100 dpi, and a vertical resolution of 100 dpi.

[0077] Step 2: Based on the formed population, configure the filter parameter groups corresponding to the individuals in the population into the hybrid filter;

[0078] Step 3: The original noisy medical image is used as input data and input into each of the hybrid filters configured with the corresponding filter parameter groups to denoise the original noisy medical image and output the original denoised medical image corresponding to each individual in the population.

[0079] Step 4: Calculate the fitness of parameters for each individual population based on the original denoised medical images corresponding to each individual population.

[0080] The fitness parameters corresponding to each individual in the population are calculated as follows:

[0081]

[0082]

[0083]

[0084]

[0085] In the formula, For the fitness parameters of an individual in the population, These are the weighting coefficients. This represents the mean square error for each individual. Peak signal-to-noise ratio for an individual. It is a structural similarity index between individuals; For noise-free raw medical images in coordinates Pixel value at that location, To denoise the medical image in coordinates Pixel value at that location, For the number of rows and columns of the image, The maximum grayscale value of the image. Original medical images, For denoising medical images, The pixel mean of the original medical image. The pixel mean of the denoised medical image. The pixel variance of the original medical image. To reduce the pixel variance of denoised medical images, The covariance between the original medical image and the denoised medical image. , is the stability constant.

[0086] Step 5: Based on the parameter fitness of each individual in the population, sort them in descending order of parameter fitness. Individuals with the highest parameter fitness in the first preset percentage are selected as global optimal solution candidates, and an attraction mechanism is triggered to drive other individuals in the population to move towards the optimal solution candidate. Individuals with the lowest parameter fitness in the second preset percentage are selected as inferior solution candidates, and a repulsion mechanism is triggered to drive other individuals in the population away from the inferior solution candidate.

[0087] The attraction mechanism in step five is as follows:

[0088] Each individual in the population Subject to global optimal solution candidates Driven by attraction, the corresponding update formula is:

[0089]

[0090] In the formula, Let be the attraction strength constant. This is a dynamic equilibrium factor that decreases linearly with the iteration number t until it reaches 0.1. A random vector between 0 and 1 A random number between 0 and 1. To switch the attraction threshold, The attractive displacement corresponds to the attraction force. In this embodiment, this application discovers a certain chaotic system in the prior art. This three-dimensional chaotic system, by introducing a hybrid driving mechanism of cross-product terms and linear terms, combined with asymmetric parameter design (such as the constant term 6(a+b) in z'), exhibits strong nonlinear dynamic characteristics, significant parameter sensitivity, and stable chaotic state generation ability while maintaining a relatively simple equation structure. It can simulate complex phenomena such as bifurcation behavior and multistable states, and is suitable for research and application in fields such as secure communication encryption, biomedical signal modeling, and complex network synchronization control. Compared with the aforementioned document, this scheme introduces dynamic singularities and positive feedback loops through nested cosine fractional nonlinear terms, resulting in a multi-level self-similar structure in the bifurcation diagram, which significantly improves the ergodicity of the solution space. Combined with Lyapunov exponent spectrum analysis, the novel system exhibits double positive Lyapunov exponents, verifying its hyperchaos characteristics, while the comparative document only has a single positive exponent. The hyperchaotic state expands the ergodicity of the solution space through high-dimensional attractors, effectively avoiding the population initialization aggregation phenomenon caused by periodic windows in traditional chaotic mapping.

[0091] To visually demonstrate the advantages of our chaotic system, we calculated the Shannon entropy of each chaotic system under optimal parameters. We used the same parameters for the number of iterations and population size to ensure the fairness of the experiment.

[0092] like Figure 5 and Figure 6 Based on the quantitative evaluation results of the two-group control experiment, the novel chaotic system proposed in this study shows a significant advantage in the Shannon Entropy (SE) index. Specifically, the Shannon entropy improvement is as follows: Under the same initial conditions and parameter configuration, the Shannon entropy value of the novel system reaches 8.999 bits, which is significantly improved compared with the comparison file 1 (baseline chaotic system, SE=4.928 bits), indicating that its phase space distribution uniformity and information randomness are significantly enhanced.

[0093] To verify the denoising effect under the same noise conditions, the following comparative experiment was conducted:

[0094] Table 1 shows a comparison of the performance indicators under the same speckle noise level. The corresponding experimental plots under the same speckle noise level are shown below. Figure 7 As shown.

[0095]

[0096] Table 1. Comparison of Indicators under Speckle Noise

[0097] Table 2 shows a comparison of the performance indicators under the same salt-and-pepper noise conditions. The corresponding experimental graphs under the same salt-and-pepper noise conditions are shown below. Figure 8 As shown.

[0098]

[0099] Table 2 Comparison of Indicators under Salt and Pepper Noise

[0100] Based on comparative experiments under the same speckle noise and salt-and-pepper noise conditions, this model demonstrates a significant advantage in edge preservation performance. The remarkable effect of LY-ARO is clearly evident in the tables and specific experimental figures above.

[0101] Step six involves determining the local search strategy corresponding to the current iteration number, and then searching for each individual in the population based on this strategy. This process optimizes each individual in the population, and step two is repeated until the iteration requirements are met. The current global optimal solution is then output as the optimal parameter set for the hybrid filter. The local search strategies in step six include Brownian motion, trigonometric functions, and random jump strategies. In this embodiment, before the first preset iteration number, the random jump strategy is primarily used, occasionally interspersed with Brownian motion or trigonometric function strategies to explore potential regions outside the direction of attraction.

[0102] Between the first and second preset number of iterations, Brownian motion and trigonometric function strategies are used as the main methods to finely adjust the parameters. When the number of iterations exceeds the second preset number, only the Brownian motion strategy is retained.

[0103] In this case, the updates of candidate solutions corresponding to the Brownian motion strategy follow a normal distribution, but the standard deviation is dynamically adjusted as follows:

[0104]

[0105]

[0106] For binary vectors, the perturbation dimension is randomly selected; Disturbance amplitude coefficient, later stage Approaching 0, it transitions to fine-tuning.

[0107] The trigonometric function strategy utilizes the periodic variations of sine and cosine functions to generate displacements, as shown in the following formula:

[0108]

[0109] In the formula, Other solutions are chosen through roulette. It is a random vector. , It is a random number. To control the amplitude, It is similar A binary vector.

[0110] The jump formula corresponding to the random jump strategy is:

[0111]

[0112] In the formula, [0,1] is a random vector. It is a binary vector, which controls the jump dimension;

[0113] In terms of strategy selection, it relies on probability switching. To determine whether to use a Brownian motion or trigonometric function strategy, otherwise use a random jump strategy:

[0114] In this embodiment, the last position is updated as follows:

[0115]

[0116] Compared to existing technologies such as BM3D, NLM, MDF, BF, MF, WF, and GF algorithms, this embodiment conducts comparative experiments on speckle noise and salt-and-pepper noise respectively:

[0117] Table 3 shows a comparison of speckle noise images, illustrating the performance of LY-ARO and existing filters under speckle noise conditions.

[0118] Comparison of noise reduction effects. Experimental data included 300 brain MRI images with speckle noise added, and the results are as follows: Figure 9 As shown, the proposed improved algorithm effectively preserves image edges and details while suppressing speckle noise, demonstrating excellent visual effects. In contrast, existing filtering algorithms are relatively coarse in contour processing, often resulting in over-smoothing of noisy areas, which in turn affects the preservation of edge structures.

[0119]

[0120] Table 3. Performance comparison of various filtering methods under speckle noise.

[0121] As shown in Table 3, our algorithm has certain advantages over BM3D, NLM, MDF, BF, MF, WF, and GF algorithms in removing image speckle noise. In contrast, existing filtering models perform poorly due to image quality degradation, processing complexity, improper edge processing, ineffective minimization of intensity variations, and inadequate noise removal. Through data comparison, our algorithm achieves a 1.28 improvement in PSNR, a 0.1416 improvement in SSIM, and a 330 reduction in MSE. This demonstrates superior image quality, edge information, and structural similarity compared to other algorithms.

[0122] Table 4 shows a comparison of the denoising effects of LY-ARO and existing filters on the dataset under the influence of salt-and-pepper noise. According to Table 4, the algorithm proposed in this study shows certain advantages over BM3D, NLM, MDF, BF, MF, WF, and GF algorithms in removing salt-and-pepper noise from images. Data comparison shows that the average PSNR was improved by 8.46, and the average MSE was reduced by 435. The results for brain MRI images with added speckle noise are shown below. Figure 10 As shown.

[0123]

[0124] Table 4. Performance comparison of various filtering methods under salt-and-pepper noise.

[0125] S2. Obtain the current medical image to be processed at the current moment, and add noise to the obtained current medical image to generate the corresponding current noisy image;

[0126] S3. Take the current noisy image as input data, input it into the mixing filter, and output the corresponding denoised medical image.

[0127] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, based on the guidance provided in this application, improve and implement this solution in combination with their own capabilities. Some typical well-known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A medical image denoising method based on 3D chaos and attraction-repulsion algorithms, characterized in that: Includes the following steps: S1. Based on historical original medical images and a preset hybrid filter parameter group optimization strategy, determine the optimal parameter group for each filter corresponding to the hybrid filter used for denoising the medical images, and configure the optimal parameter group into the corresponding filter; the preset hybrid filter parameter group optimization strategy includes a 3D-LY chaotic mapping strategy for initializing the population corresponding to the hybrid filter parameter group and an improved attraction-repulsion optimization strategy for optimizing the initialized population. S2. Obtain the current medical image to be processed at the current moment, and add noise to the obtained current medical image to generate the corresponding current noisy image; S3. The current noisy image is used as input data and fed into the mixing filter to output the corresponding denoised medical image; The execution process of the 3D-LY chaotic mapping strategy for population initialization is as follows: Based on the number of filter types in the hybrid filter, a corresponding three-dimensional chaotic model is constructed; the three-dimensional chaotic model is: In the formula, , , These are the system state variables in the corresponding three-dimensional chaotic model. , These are the parameters of the three-dimensional chaotic model; Based on the system state variables in the three-dimensional chaotic model, the filter parameters corresponding to each filter in the hybrid filter are mapped to one of the system state variables respectively; The three-dimensional chaotic model is run and iterated multiple times, and the system state variables corresponding to each iteration are recorded to form a corresponding chaotic sequence matrix; each row of the chaotic sequence matrix corresponds to a filter parameter group. Normalize each row of chaotic sequence in the formed chaotic sequence matrix to form the individuals of the corresponding initial population, and merge the individuals to form the corresponding initial population. The improved attraction-repulsion optimization strategy is executed as follows: Step 1, acquire historical original medical images and add noise to the original medical images to generate corresponding original noisy medical images. Step 2: Based on the formed population, configure the filter parameter groups corresponding to the individuals in the population into the hybrid filter; Step 3: The original noisy medical image is used as input data and input into each of the hybrid filters configured with the corresponding filter parameter groups to denoise the original noisy medical image and output the original denoised medical image corresponding to each individual in the population. Step 4: Calculate the fitness of parameters for each individual population based on the original denoised medical images corresponding to each individual population. Step 5: Based on the parameter fitness of each individual in the population, sort them in descending order of parameter fitness. Individuals with the highest parameter fitness in the first preset percentage are selected as global optimal solution candidates, and an attraction mechanism is triggered to drive other individuals in the population to move towards the optimal solution candidate. Individuals with the lowest parameter fitness in the second preset percentage are selected as inferior solution candidates, and a repulsion mechanism is triggered to drive other individuals in the population away from the inferior solution candidate. Step 6: Based on the current iteration number, determine the local search strategy corresponding to the iteration number, and search for each individual in the population based on the determined local search strategy to form new individuals after optimization of each individual in the population. Then, repeat step 2 until the iteration requirements are met, and output the current global optimal solution as the optimal parameter set of the filter corresponding to the hybrid filter.

2. The medical image denoising method based on 3D chaos and attraction-repulsion algorithm according to claim 1, characterized in that: The fitness parameters corresponding to each individual in the population are calculated as follows: In the formula, For the fitness parameters of an individual in the population, These are the weighting coefficients. This represents the mean square error for each individual. Peak signal-to-noise ratio for an individual. It is a structural similarity index between individuals; For noise-free raw medical images in coordinates Pixel value at that location, To denoise the medical image in coordinates Pixel value at that location, For the number of rows and columns of the image, The maximum grayscale value of the image. Original medical images, For denoising medical images, The pixel mean of the original medical image. The pixel mean of the denoised medical image. The pixel variance of the original medical image. To reduce the pixel variance of denoised medical images, The covariance between the original medical image and the denoised medical image. , is the stability constant.

3. The medical image denoising method based on 3D chaos and attraction-repulsion algorithm according to claim 2, characterized in that: The attraction mechanism in step five is as follows: Each individual in the population Subject to global optimal solution candidates Driven by attraction, the corresponding update formula is: In the formula, Let be the attraction strength constant. This is a dynamic equilibrium factor that decreases linearly with the iteration number t until it reaches 0.

1. A random vector between 0 and 1 A random number between 0 and 1. To switch the attraction threshold, The attractive displacement corresponding to the attractive force.

4. The medical image denoising method based on 3D chaos and attraction-repulsion algorithm according to claim 3, characterized in that: The local search strategies in step six include Brownian motion strategy, trigonometric function strategy, and random jump strategy.

5. A medical image denoising system based on 3D chaos and attraction-repulsion algorithms, characterized in that: The medical image denoising method based on 3D chaos and attraction-repulsion algorithm as described in any one of claims 1 to 4.

Images