Medical image denoising method based on 3D chaotic mapping dung beetle algorithm

Abstract

The application relates to the technical field of image processing, and discloses a medical image denoising method based on a 3D chaotic mapping algorithm, which comprises the following steps: step 1, inputting an original medical image to be processed, performing noise adding processing on the original medical image, and generating a noisy image; step 2, inputting the noisy image into a hybrid filter used for denoising the image; step 3, calculating optimal parameter combinations of each filter in the hybrid filter by adopting a dung beetle optimization algorithm, and configuring the hybrid filter by adopting the optimal parameter combinations; and step 4, performing denoising processing on the noisy image by using the configured hybrid filter, and generating a denoised image; in step 3, the distribution positions of a population are initialized by adopting a 3D chaotic mapping; and the technical problem of poor medical image denoising effect in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a medical image denoising method based on a 3D chaos mapping dung beetle algorithm. Background Art

[0002] Medical images, such as X-rays, CT scans, and MRIs, are crucial for doctors to diagnose diseases. However, due to limitations in imaging equipment and individual patient differences, medical images often contain varying degrees of noise. This noise can obscure important pathological information, leading to misdiagnosis or missed diagnosis. Therefore, image denoising technology, which improves the clarity and signal-to-noise ratio of medical images, is crucial for enhancing diagnostic accuracy and reliability.

[0003] Many existing filtering techniques reduce noise in images by smoothing or blurring. However, these techniques often lead to the loss of complex details in the image, especially in edge and texture areas. Therefore, finding a balance between noise reduction and retaining important image features is a challenging task. The parameter configuration of the filter is the key to achieving this balance. The choice of parameters directly affects the filtering effect, including the degree of noise removal and the retention of image details. In order to automatically calculate the optimal parameters of the filter, existing technologies use meta-heuristic algorithms, such as genetic algorithms and particle swarm optimization algorithms. These algorithms search for the optimal solution by simulating natural processes (such as genetic evolution, group behavior, etc.).

[0004] In metaheuristic algorithms, the distribution of the initial population significantly affects the accuracy and convergence speed of the algorithm. A reasonable initial population distribution can accelerate the algorithm's convergence process and improve search efficiency; however, an unreasonable initial population distribution can cause the algorithm to become trapped in a local optimum and reduce search accuracy. To address the issue of uncertain initial population distribution, the initial positions of the population should be evenly distributed throughout the search space to ensure diversity and traversability. However, simply generating random position vectors cannot guarantee efficient traversal and diversity.

[0005] A commonly used method in existing technologies is chaos mapping. Chaos mapping is a nonlinear dynamic system with pseudo-randomness and ergodic properties that can generate time series with complex behavior. Common chaos mapping methods include tent chaos mapping, logistic chaos mapping, and sinusoidal chaos mapping. Although the use of chaos mapping allows algorithms to escape local optimal solutions, it still suffers from the serious drawback of randomness. That is, while the initial population positions are diverse, they lack a certain regularity. This results in large fluctuations and uncertainty in the algorithm's search process, affecting search efficiency and accuracy. Summary of the Invention

[0006] The present invention aims to provide a medical image denoising method based on a 3D chaotic mapping dung beetle algorithm to solve the technical problem of poor medical image denoising effect in the prior art.

[0007] To achieve the above object, the present invention adopts the following technical solution: a medical image denoising method based on a 3D chaotic mapping dung beetle algorithm, comprising the following steps:

[0008] Step 1: inputting an original medical image to be processed, performing noise processing on the original medical image, and generating a noisy image;

[0009] Step 2: input the noisy image into a hybrid filter for denoising the image;

[0010] Step 3, using the dung beetle optimization algorithm to calculate the optimal parameter combination of each filter in the hybrid filter, and using the optimal parameter combination to configure the hybrid filter;

[0011] Step 4: De-noising the noisy image using the configured hybrid filter to generate a de-noised image;

[0012] Among them, in step 3, the distribution position of the population is initialized using 3D chaotic mapping, and the 3D chaotic mapping is:

[0013]

[0014] Among them, x, y, z are system state variables, a=-0.6, c=-10, 1≤b≤7, and the initial values ​​are set to (0.1, 0, 0).

[0015] The principle and advantages of this scheme are as follows: A raw medical image to be processed is input and subjected to noise processing to generate a noisy image. By artificially adding a certain amount of noise (such as Gaussian noise or salt and pepper noise) to the image, the noise components in the image are made more prominent and easier to identify. The noisy image is then input into a hybrid filter. This hybrid filter is composed of multiple filters, aiming to improve the denoising effect by combining the advantages of different filters.

[0016] The dung beetle optimization algorithm is used to calculate the optimal parameter combination for each filter in the hybrid filter. A 3D chaotic map is used to initialize the distribution of the dung beetle population. The pseudo-random and ergodic properties of the 3D chaotic map enable the generation of time series with complex behaviors, thereby ensuring the diversity and ergodicity of the initial population. The 3D chaotic map ensures that the initial positions of the population are evenly distributed throughout the search space, reducing the possibility of falling into local optima.

[0017] This paper proposes a novel 3D chaotic mapping method that is highly sensitive to initial conditions and can generate random sequences with unique properties. Even small differences in initial conditions can lead to completely different evolutionary trajectories, thus ensuring a uniform distribution of the population and reducing the possibility of being trapped in local optima.

[0018] The dung beetle optimization algorithm, combined with 3D chaotic mapping, exhibits greater flexibility within the global search range. By iteratively searching for the optimal filter parameter combination, this method adapts to the complex characteristics of medical images and effectively addresses various types of medical noise. This method effectively preserves key image features, such as edges and texture information, during the denoising process, thereby improving the visual quality and diagnostic value of medical images.

[0019] The 3D chaotic mapping algorithm ensures uniformity when initializing the dung beetle population, effectively reducing convergence time, alleviating the computational burden, and accelerating processing. By reducing defects such as noise and artifacts, this method improves the quality of medical images and helps reduce data redundancy, thereby simplifying the storage and transmission of medical images.

[0020] Preferably, as an improvement, in step 3, the optimal value of b is 3.7.

[0021] The beneficial effect of this improvement is that, through a detailed analysis of the parameter b of the 3D chaotic mapping using the Lyapunov exponent spectrum and bifurcation diagrams, it was found that when b = 3.7, the curvature of the Lyapunov exponent reaches its maximum value, indicating that the system has the strongest chaotic characteristics and the best ergodicity under this parameter. Therefore, determining the optimal value of parameter b as 3.7 can significantly improve the efficiency and effectiveness of the 3D chaotic mapping when initializing the dung beetle population, thereby enhancing the global search capability and convergence speed of the dung beetle optimization algorithm.

[0022] Preferably, as an improvement, in step 2, the hybrid filter includes: a Gaussian filter, a median filter, a mean filter, a non-local mean filter and a sharpening filter.

[0023] The beneficial effect of this improvement is that the Gaussian filter, median filter, mean filter, non-local mean filter, and sharpening filter are combined into a hybrid filter, which can fully leverage the advantages of each filter to achieve a more comprehensive denoising effect. The Gaussian filter effectively reduces Gaussian noise; the median filter excels at removing salt and pepper noise; the mean filter slightly reduces image sharpness but can simplify image structure in some cases; the non-local mean filter preserves more detail; and the sharpening filter enhances image edges and details. This combination can cope with various types of noise while preserving important image features as much as possible.

[0024] Preferably, as an improvement, the dung beetle algorithm includes a rolling ball behavior. During the rolling behavior, the calculation formula for updating the dung beetle position is:

[0025]

[0026] x i (t+1) is the position of the i-th dung beetle at the t-th iteration, k∈(0,0.2] is a constant value indicating the deflection coefficient, b∈(0,1) is a constant; α is -1 or 1, X w is the global worst position, Δx is the degree of change in light intensity;

[0027] When α=1, it indicates that the natural environment does not affect the original direction, and when α=-1, it indicates that the original direction is deviated from.

[0028] The beneficial effect of this improvement is that by incorporating a deflection coefficient and light intensity variations into the rolling behavior, the dung beetle can adjust its direction based on the current environment (i.e., light intensity). When the environment is stable (α = 1), the dung beetle maintains its original rolling direction; when the environment changes (α = -1), the dung beetle deviates from its original rolling direction. This mechanism increases the diversity of the dung beetle's search paths, helps avoid being trapped in local optimal solutions, and improves its global search capabilities.

[0029] Preferably, as an improvement, the dung beetle algorithm includes dancing behavior. During the dancing behavior, the calculation formula for updating the position of the dung beetle is:

[0030] x i (t+1)=x i (t)+tan(θ)|x i (t)-x i (t-1)|;

[0031] |x i (t)-x i (t-1)| is the difference between the position of the i-th dung beetle at the t-th iteration and its position at the t-1-th iteration; θ∈[0,π] is the deflection angle; when θ is equal to 0,π / 2,π, the dung beetle does not change its position.

[0032] The beneficial effect of this improvement is that during the dancing behavior, the dung beetle updates its position by calculating the position difference and deflection angle. When the position difference and deflection angle meet certain conditions, the dung beetle will not change its position, which helps maintain stability during the search process. At the same time, this mechanism also allows the dung beetle to make small position adjustments when necessary to explore new search areas. Therefore, the dancing behavior can balance global and local search, improving the algorithm's convergence speed and accuracy.

[0033] Preferably, as an improvement, the dung beetle algorithm includes a breeding behavior, and in the breeding behavior, the calculation formula for the egg-laying area is:

[0034] Lb * =max(X * ×(1-R), Lb), Ub * =min(X * ×(1+R),Ub);

[0035] Lb * , Ub * are the lower and upper limits of the spawning area, Lb and Ub are the lower and upper limits of the search space, X * is the optimal position in the current population, and the inertia weight R = 1-t / T max , T max The maximum number of iterations during algorithm iteration;

[0036] The calculation formula for the position of the incubation fecal ball is:

[0037] B i (t+1)=X * +b1×(B i (t)-Lb * )+b2×(B i (t)-Ub * );

[0038] B i (t) is the position of the i-th incubated fecal ball at the t-th iteration, b1 and b2 represent two independent random vectors of size 1×D, and D is the dimension of the optimization problem.

[0039] The beneficial effect of this improvement is that during reproduction, the oviposition area and the location of the brooding balls are calculated, enabling the renewal and evolution of the dung beetle population. The oviposition area is determined based on the optimal location within the current population and the upper and lower bounds of the search space, ensuring that the newly generated brooding balls are located within the search space. The calculation of the brooding ball location combines randomness with the dimensionality of the optimization problem, increasing the diversity of the population. This reproduction mechanism helps maintain the vitality and diversity of the population and improves the algorithm's global search capabilities.

[0040] Preferably, as an improvement, the dung beetle algorithm includes foraging behavior, and in the foraging behavior, the calculation formula for the optimal foraging area is:

[0041] Lb b =max(X b ×(1-R), Lb), Ub b =min(X b ×(1+R),Ub);

[0042] X b is the global optimal position, Lb b 、Ub bare the lower and upper limits of the optimal foraging area;

[0043] The calculation formula for the position update of the dung beetle is:

[0044] x i (t+1)=x i (t)+C1×(x i (t)-Lb b )+C2×(x i (t)-Ub b );

[0045] x i (t) represents the position information of the i-th dung beetle at the t-th iteration, C1 represents a random number following the normal distribution, and C2 represents a random vector belonging to (0,1).

[0046] The beneficial effect of this improvement is that during foraging, the optimal foraging area and the position update formula for the dung beetle are calculated to guide the dung beetle towards the global optimal position. The optimal foraging area is determined based on the global optimal position and the upper and lower bounds of the search space, providing the dung beetle with a clear search direction. The dung beetle's position update combines the randomness of a normal distribution with random vectors, increasing the diversity of the search path. This foraging mechanism accelerates the algorithm's convergence process and improves search efficiency.

[0047] Preferably, as an improvement, the dung beetle algorithm includes stealing behavior. During the stealing behavior, the calculation formula for the position update of the thief dung beetle is:

[0048] x i (t+1)=X b +S×g×(|x i (t)-X * |+|x i (t)-X b |);

[0049] where x i (t) represents the position of the i-th thief in the t-th iteration, g is a random vector of size 1×D, which obeys the normal distribution, S represents a constant, X b To compete for the best position for food, X * is the optimal position in the current population.

[0050] The beneficial effect of this improvement is that during their stealing behavior, the thieving dung beetles mimic the behavior of the optimal position within the current population by calculating a position update formula. This mechanism introduces elements of competition and cooperation, enabling the dung beetle population to learn from each other and make progress together during the search process. Furthermore, because the thieving dung beetles' position updates combine the properties of randomness and normal distribution, they improve the algorithm's search capabilities while maintaining population diversity. Stealing behavior helps the algorithm escape local optima and enhance its global search capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a denoising flowchart of an embodiment of the present invention.

[0052] Figure 2 1 is the Lyapunov exponent spectrum and bifurcation diagram under the change of parameter b in an embodiment of the present invention.

[0053] Figure 3 This is the phase diagram of the chaotic map according to the embodiment of the present invention.

[0054] Figure 4 This is a comparison diagram of the speckle denoising effect of an embodiment of the present invention.

[0055] Figure 5 This is a comparison diagram of the Poisson denoising effect of an embodiment of the present invention.

[0056] Figure 6 This is a comparison diagram of the salt and pepper denoising effect of an embodiment of the present invention.

[0057] Figure 7 This is a comparison diagram of Gaussian denoising effects according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following is further described in detail through specific implementation methods:

[0059] Example

[0060] Basically as attached Figure 1 As shown, the medical image denoising method based on the 3D chaos mapping dung beetle algorithm includes the following steps:

[0061] Step 1: inputting an original medical image to be processed, performing noise processing on the original medical image, and generating a noisy image;

[0062] Step 2: input the noisy image into a hybrid filter for denoising the image;

[0063] Step 3, using the dung beetle optimization algorithm to calculate the optimal parameter combination of each filter in the hybrid filter, and using the optimal parameter combination to configure the hybrid filter;

[0064] Step 4: De-noising the noisy image using the configured hybrid filter to generate a de-noised image;

[0065] Among them, in step 3, the distribution position of the population is initialized using 3D chaotic mapping, and the 3D chaotic mapping is:

[0066]

[0067] Among them, x, y, z are system state variables, and a, b, c are system parameters; assume that the system parameters a = -0.6, c = -10, 1≤b≤7, and the initial values ​​are set to (0.1, 0, 0).

[0068] The optimal value of parameter b is determined by using the Lyapunov index spectrum and bifurcation diagram. The initial conditions of the system are set to (0.1, 0, 0), and the parameters a = -0.6 and c = -10; the following is obtained: Figure 2 As shown, the upper figure is the Lyapunov index diagram, and the lower figure is the bifurcation diagram. When 1≤b≤7, the system is always in a chaotic state.

[0069] The curvature of the Lyapunov exponent reaches its maximum value when b = 3.7, so the optimal value of parameter b is determined to be 3.7. Figure 3 As shown in the figure, the chaotic map phase diagram is set with system parameters a = -0.6, b = 3.7, c = -10, and the system initial value is determined to be (0.1, 0, 0). Among them, (a) is 3D, (b) is the xy plane, (c) is the xz plane, and (d) is the yz plane.

[0070] The hybrid filter used in step 2 includes: a Gaussian filter (GF), a median filter (MDF), a mean filter (MF), a non-local mean filter (NLM) and a sharpening filter.

[0071] A Gaussian filter (GF) smoothes an image by convolving it with a Gaussian function, thereby reducing noise and detail. Based on a set standard deviation, o, a Gaussian-distributed weight matrix (i.e., a Gaussian kernel) is generated. The Gaussian kernel is convolved with the image. For each pixel in the image, the pixel values ​​in its neighborhood are multiplied by the weight corresponding to the Gaussian kernel, and the sum is calculated to obtain a new, smoothed pixel value. Through this process, the Gaussian filter can effectively reduce high-frequency noise in the image while preserving the overall image structure. The expression formula for the Gaussian filter is as follows.

[0072]

[0073] Where G(x, y) represents the distance between pixels, and σ represents the standard deviation of the Gaussian distribution. A Gaussian filter smooths an image by applying a Gaussian kernel. This smoothing effect is achieved by taking a weighted average of the values ​​of each pixel's neighbors, with the weights determined by the Gaussian function. This reduces the effects of noise on the image.

[0074] Median filter (MDF), the working principle of the median filter is to sort the neighborhood pixel values ​​of each pixel in the image, and then take the median as the new value of the pixel, so as to achieve the effect of removing noise, especially having a good suppression effect on salt and pepper noise. For a pixel I (x, y), its neighborhood is N (x, y) (for example, a 3×3 window), the output of the median filter I filtered (x, y) is expressed as:

[0075] I filtered (x, y)median=I(i,j)|(i,j)∈N(x,y);

[0076] Where median represents the median operation; N(x, y) is the neighborhood of pixel I(x, y). The median filter removes salt and pepper noise or other noise by sorting the pixel values ​​of each pixel and its neighborhood and replacing the original pixel value with the median value.

[0077] Mean filter (MF) is a filter that smoothes and reduces noise by averaging the values ​​of each pixel in the image and its neighborhood. It can effectively reduce random noise in the image, but it may also cause blurring of image details. For a pixel I (x, y), its neighborhood is N (x, y) (for example, a 3 × 3 window), the output of the mean filter I filtered (x, y) is expressed as:

[0078]

[0079] Where |N(x, y)| is the number of pixels in the neighborhood. I(i, j) is the value of the pixel in the neighborhood N(x, y). A mean filter reduces the sharpness and clarity of an image by smoothing out details and noise. This is because each pixel value becomes the average of all the pixel values ​​in its neighborhood, reducing the differences between pixels.

[0080] Non-local mean filter (NLM), Non-local mean filter (NLM) performs smoothing and denoising by considering similar pixel blocks in the image, not just the local neighborhood. For each pixel, NLM not only uses the pixels in its local neighborhood, but also looks for small blocks similar to the pixel in the entire image (usually a window of fixed size), and the pixel values ​​of these similar blocks will be used to calculate the new pixel value. The similarity between similar pixel blocks is calculated by Gaussian weighting. The higher the similarity, the greater the weight. In the end, the filtered pixel value is the weighted average of the pixel values ​​in these similar blocks. The similarity is usually calculated based on Euclidean distance (or other metrics). The smaller the distance, the more similar the pixel blocks are, and the greater the weight. Given an image I, the value I after non-local mean filtering of a pixel I(p) in the image filtered (p) is represented by the following formula:

[0081]

[0082] Where p is a pixel position in the image; Ω is the region in the image where similar blocks are searched; I(q) is the pixel value at position q in the image; and w(p,q) is the similarity weight between the pixel blocks at positions p and q. Non-local means filtering can better preserve image details. By calculating similarity and performing a weighted average, it can reduce the effects of noise, resulting in a clearer and more natural image.

[0083] Sharpening filter: Sharpening filter can make the image after the previous filters look clearer and sharper, and enhance the details and features of the image. This paper uses the sharpening method of high-lift filtering, and the formula is as follows:

[0084] I sharpened (x,y)=I(x,y)+α·(I(x,y)-I blurred (x, y));

[0085] I(x, y) is the original image. blurred (x, y) is the low-pass filtered image. α is the coefficient that controls the sharpening strength, typically greater than 1. The sharpening filter applies a high-pass filter to extract high-frequency components from the image, performs a second-order derivative of the Laplacian operator on the image to enhance edges, and then adds the original image to the high-pass filtered image, thereby enhancing image detail and edges.

[0086] Step 1 performs noise processing. By artificially adding a certain amount of noise (such as Gaussian noise or salt and pepper noise) to the image, the noise components in the image are made more obvious and easier to identify. This enables the hybrid filter to exhibit better denoising performance when faced with real noise. Because the artificial noise is added, the hybrid filter has learned how to identify and process these noise components during the training process, making it more effective in removing them in practical applications.

[0087] Hybrid filters, such as non-local means filters, use the similarities between pixels in an image to estimate and restore the original pixel values ​​when removing noise. By adding noise first, the diversity of pixel values ​​in the image can be increased, making the hybrid filter more adaptable to different types of noise during application and enhancing its robustness.

[0088] Among hybrid filters, non-local mean filters can better protect image edges when removing noise, and noise addition processing can further optimize the performance of hybrid filters, enabling them to better preserve image edge features while removing noise.

[0089] The Dung Beetle Optimization (DBO) algorithm in step 3 is an optimization algorithm inspired by the natural behavior of dung beetles. The algorithm includes rolling, breeding, foraging, stealing, and dancing behaviors.

[0090] Rolling behavior: In the algorithm, it is assumed that the intensity of the sun also affects the path of the dung beetle. During the rolling process, the position of the dung beetle is updated as follows:

[0091]

[0092] Here x i (t+1) is the position of the i-th dung beetle at the t-th iteration, k∈(0,0.2] represents a constant value indicating the deflection coefficient, b∈(0,1) is a constant; α is a natural coefficient assigned to -1 or 1, X w It is used to indicate the global worst position, and Δx is used to indicate the degree of change in light intensity.

[0093] α is probabilistically set to 1 or -1, with α = 1 indicating that the natural environment does not affect the original direction, while α = -1 indicates deviation from the original direction. A higher value of Δx indicates a weaker light source, encouraging the dung beetle to avoid that location. This allows the entire search space to be explored as thoroughly as possible during the optimization process, reducing the likelihood of falling into a local optimum.

[0094] Dancing behavior: When a dung beetle encounters an obstacle and cannot move forward, it needs to dance to locate itself and find a new route. To simulate dancing behavior, a tangent function is used to obtain the new rolling direction, which is in the interval [0, π]. Once the dung beetle successfully determines the new direction, it should continue to roll the dung ball backward. Therefore, the dung beetle's position update method when rolling the dung ball is:

[0095] x i (t+1)=x i (t)+tan(θ)|x i (t)-x i (t-1)|;

[0096] | x i (t)-x i (t-1)| is the difference between the position of the i-th dung beetle at iteration t and its position at iteration t-1. Therefore, the dung beetle's position update is closely related to current and historical information. Here, θ∈|0,π] is the deflection angle. Note that when θ equals 0,π / 2,π, the dung beetle does not change its position.

[0097] Reproduction behavior: When the dung ball is rolled back home, in order to provide a safe environment for the offspring, a suitable place for laying eggs is selected. A boundary selection strategy is introduced to simulate the area where the female insect lays eggs, which is defined as follows:

[0098] Lb * =max(X * ×(1-R), Lb), Ub * =min(X * ×(1+R),Ub);

[0099] Among them Lb * , Ub * are the lower and upper limits of the spawning area, Lb and Ub are the lower and upper limits of the search space, X * is the optimal position in the current population, and the inertia weight R = 1-t / T max , where T max The maximum number of iterations of the algorithm.

[0100] Once the egg-laying area is determined, the female insect will choose a brooding ball in that area to lay her eggs. For the DBO algorithm, each female insect will produce only one egg in each iteration, which means there is only one solution. The boundaries of the egg-laying area are dynamically changing to prevent the algorithm from falling into a local optimum, which is mainly determined by the inertia weight R. Therefore, the location of the brooding ball is also dynamic during the iteration process and is defined as follows:

[0101] B i (t+1)=X * +b1×(Bi (t)-Lb * )+b2×(B i (t)-Ub * );

[0102] Among them B i (t) is the position of the i-th brooding ball at iteration t, b1 and b2 are two independent random vectors of size 1 × D, where D is the dimension of the optimization problem. Note that the position of the brooding ball is strictly restricted to a certain range, namely the spawning area.

[0103] Foraging behavior: When the baby dung beetles hatch successfully, they will run out in groups to forage, so it is necessary to establish the best foraging area to guide them.

[0104] Lb b =max(X b ×(1-R), Lb), Ub b =min(X b ×(1+R),Ub);

[0105] where X b is the global optimal position, Lb b 、Ub b are the lower and upper limits of the optimal foraging area. After determining the area, we can define the location update method of the dung beetle:

[0106] x i (t+1)=x i (t)+C1×(x i (t)-Lb b )+C2×(x i (t)-Ub b );

[0107] Here x i (t) represents the position information of the i-th dung beetle at the t-th iteration, C1 represents a random number following the normal distribution, and C2 represents a random vector belonging to (0,1).

[0108] Thief behavior: Some dung beetles, called thieves, steal dung balls from other dung beetles. b is the best food source. Assume that X b Left and right represent the best positions for grabbing food. During the iteration, the thief's position information is updated and can be described as follows:

[0109] x i (t+1)=X b +S×g×(|x i (t)-X*|+|x i (t)-X b |);

[0110] where x i (t) represents the position of the i-th thief in the t-th iteration, g is a random vector of size 1×D, which obeys the normal distribution, S represents a constant, X b Left and right represent the best position to compete for food, X * is the optimal position in the current population.

[0111] To verify the performance of this solution, this example uses four brain slices of consistent quality but different types to compare the denoising effect. The brain slice data comes from brain MRI images published by Kaggle.

[0112] (Source: https: / / www.kaggle.com / datasets / ashfakyeafi / brain-mri-images).

[0113] This dataset has a size of 496×248, a bit depth of 24, a horizontal resolution of 100 dpi, and a vertical resolution of 100 dpi.

[0114] Multiple indicators are used for evaluation, including mean square error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), edge preservation index (EPI), and ENR index.

[0115] The mean square error (MSE) shows the difference between two images and is defined as:

[0116]

[0117] Among them, I0 is the noise-free original image, I n is the noise image, k and l are the number of rows and columns respectively.

[0118] Peak signal-to-noise ratio (PSNR) reflects the quality of an image estimated from the original image. It gives the objective difference between two images and can be defined as the ratio of peak signal power to average noise power:

[0119]

[0120] Where N is the number of gray levels in the image.

[0121] The Structural Similarity Index (SSIM), a computational metric used to understand the similarity between two images (x, y), is defined as:

[0122]

[0123] Where c1 and c2 are constants, μ xis the mean value of x, μ y is the mean value of y, and are the variances of x and y, σ xy represents the covariance of x and y.

[0124] The edge preservation index (EPI) is an indicator used to measure the degree of edge information preservation after image denoising or filtering. It is defined as:

[0125]

[0126] in, represents the edge strength of the original image at the i-th pixel, It represents the edge strength of the denoised or processed image at the i-th pixel, and n is the total number of pixels in the image.

[0127] The ENR index is used to measure the relationship between the edge information and noise of an image. It is defined as:

[0128]

[0129]

[0130] in, and are the gradients of the image in the x and y directions, respectively. m and n are the number of rows and columns of the image, respectively. N is the total number of pixels in the image.

[0131] The image denoising algorithm proposed in this embodiment is referred to as the GBA algorithm. The denoising effects of the GBA algorithm are compared with those of existing filters. The existing filters used for comparison are: a three-dimensional block matching filter (BM3D), a non-local means filter (NLM), a multi-delay block frequency domain adaptive filter (MDF; a method that divides the original multi-order filter into K equally divided sub-blocks and performs adaptive filtering on each sub-block of length N), a bilateral filter (BF), a mean filter (MF), a Wiener filter (WF; an optimal estimator based on the minimum mean square error criterion), and a Gaussian filter (GF).

[0132] 1. Speckle Noise Image Comparison

[0133] The brain MRI images of the central axial slice of the brain were selected to add speckle noise, and the GBA algorithm and the existing filter were used to perform noise reduction. The denoising results are shown in the attached figure. Figure 4As shown in Figure 1, (a) is the original MRI image, (b) is the image after adding speckle noise, (c) is the denoised image generated by GBA, (d) is the denoised image generated by BM3D, (e) is the denoised image generated by NLM, (f) is the denoised image generated by MDF, (g) is the denoised image generated by BF, (h) is the denoised image generated by MF, (i) is the denoised image generated by WF, and (j) is the denoised image generated by GF. The small rectangle that enlarges the ROI shows the large matrix part.

[0134] GBA effectively preserves image edges and details while suppressing speckle noise, delivering excellent visual results. In contrast, existing filtering algorithms are relatively rough in contour processing, often resulting in over-smoothing of noisy areas, which in turn affects the preservation of edge structures.

[0135] Table 1 shows the performance comparison of various filtering methods under speckle noise.

[0136]

[0137]

[0138] As shown in Table 1, the GBA algorithm achieved a PSNR improvement of 1.28, an SSIM improvement of 0.1416, a MSE reduction of 0.0005, an ENR improvement of 6.69, and an EPI improvement of 4.5705, which shows that it has superior image quality, edge information, and structural similarity compared to other algorithms.

[0139] 2. Poisson Noise Image Comparison

[0140] The brain MRI images of the coronal slices of the brain were selected and Poisson noise was added. The GBA algorithm and the existing filter were used to perform noise reduction. The denoising results are shown in the attached figure. Figure 5 As shown in the figure, (a) is the original MRI image, (b) is the image after adding Poisson noise, (c) is the denoised image generated by GBA, (d) is the denoised image generated by BM3D, (e) is the denoised image generated by NLM, (f) is the denoised image generated by MDF, (g) is the denoised image generated by BF, (h) is the denoised image generated by MF, (i) is the denoised image generated by WF, and (j) is the denoised image generated by GF. The small rectangle of the magnified ROI shows the large matrix part; the area indicated by the circle and arrow is the focus area.

[0141] GBA as a whole shows better Poisson noise removal effect than traditional methods. Compared with existing filtering methods, GBA algorithm can clearly retain texture features and enhance edge information when suppressing Poisson noise in images. Figure 5As shown in the circled area indicated by the arrow, GBA performs well in noise suppression and contour detail preservation, significantly outperforming other algorithms, fully demonstrating its superiority in image processing.

[0142] Table 2 shows the performance comparison of various filtering methods under Poisson noise.

[0143] Metrics GBA BM3D NLM MDF BF MF WF GF PSNR 26.07 23.90 22.17 15.14 17.14 20.56 17.7 21.79 SSIM 0.8028 0.6949 0.7762 0.3974 0.5004 0.5996 0.5954 0.6951 MSE 0.0036 0.0042 0.0061 0.0306 0.0193 0.0088 0.017 0.0066 ENR 9.3954 6.8811 5.7370 1.1176 1.7792 3.9010 2.0165 5.2625 EPI 11.4858 4.4288 2.1489 2.4154 2.6592 6.6250 3.9804 8.2830

[0144] As shown in Table 2, the GBA algorithm achieved a PSNR improvement of 2.17, an SSIM improvement of 0.1079, a MSE reduction of 0.0006, an ENR improvement of 2.51, and an EPI improvement of 3.2028. This demonstrates its ability to preserve edge information well, resulting in relatively low noise levels and overall excellent image quality.

[0145] 3. Salt and Pepper Noise Image Comparison

[0146] The brain MRI images of the central slice were selected to add salt and pepper noise, and the GBA algorithm and the existing filter were used to perform noise reduction. The denoising results are shown in the attached figure. Figure 6 As shown in the figure, (a) is the original MRI image, (b) is the image after adding salt and pepper noise, (c) is the denoised image generated by GBA, (d) is the denoised image generated by BM3D, (e) is the denoised image generated by NLM, (f) is the denoised image generated by MDF, (g) is the denoised image generated by BF, (h) is the denoised image generated by MF, (i) is the denoised image generated by WF, and (j) is the denoised image generated by GF.

[0147] The BM3D, NLM, BF, MF, WF, and GF algorithms all demonstrated similar performance in removing salt and pepper noise. While the MDF algorithm performed relatively well in dealing with salt and pepper noise, some subtle noise remained incompletely removed, potentially hindering accurate lesion identification during clinical diagnosis. In contrast, the GBA algorithm, building on the MDF algorithm, performs a more in-depth denoising process across the entire brain region, effectively preserving contours and details while significantly eliminating clutter. This improvement can significantly enhance medical staff's ability to identify lesions and provide a better understanding of a patient's condition.

[0148] Table 3 shows the performance comparison of various filtering methods under salt and pepper noise.

[0149] Metrics GBA BM3D NLM MDF BF MF WF GF PSNR 26.41 16.27 15.25 17.95 8.7 13.87 14.55 14.41 SSIM 0.7993 0.2534 0.3674 0.8061 0.0541 0.1864 0.3256 0.2827 MSE 0.0024 0.0240 0.0299 0.0160 0.1350 0.0411 0.0350 0.0362 ENR 14.1934 1.187 1.1479 2.0932 0.2187 0.8567 0.9762 0.9685 EPI 9.2264 0.7432 0.5662 4.0428 1.7794 2.5710 1.9767 2.6669

[0150] As shown in Table 3, the data comparison shows that PSNR is improved by 8.46, MSE is reduced by 0.0136, ENR is improved by 11.2614, and EPI is improved by 5.1836. However, the SSIM index is reduced by 0.0068 compared with MDF. Although the MDF algorithm is slightly less effective in removing salt and pepper noise, it has certain advantages in preserving the details of the original image.

[0151] 4. Gaussian noise image comparison

[0152] The brain MRI images of the central axial slice were selected to add Gaussian noise, and the GBA algorithm and the existing filter were used to perform noise reduction. The denoising results are shown in the attached figure. Figure 7 As shown in the figure, (a) is the original MRI image, (b) is the image after adding Gaussian noise, (c) is the denoised image generated by GBA, (d) is the denoised image generated by BM3D, (e) is the denoised image generated by NLM, (f) is the denoised image generated by MDF, (g) is the denoised image generated by BF, (h) is the denoised image generated by MF, (i) is the denoised image generated by WF, and (j) is the denoised image generated by GF. The small rectangle that enlarges the ROI shows the large matrix part.

[0153] The BM3D, NLM, BF, MF, WF, and GF algorithms show little difference in edge detail and spatial structure between images when processing Gaussian noise. Although the BM3D algorithm's denoising capabilities are slightly improved compared to traditional methods, its performance with Gaussian noise is inferior to that of the GBA algorithm. Clearly, the GBA algorithm demonstrates significant advantages in suppressing Gaussian noise while effectively preserving edge details in the image.

[0154] Table 4 shows the performance comparison of various filtering methods under Gaussian noise

[0155] Metrics GBA BM3D NLM MDF BF MF WF GF PSNR 23.51 20.32 19.74 18.95 17.76 18.30 18.60 19.01 SSIM 0.5679 0.3822 0.4391 0.3667 0.2622 0.3258 0.3612 0.3985 MSE 0.0045 0.0094 0.0106 0.0127 0.0167 0.0148 0.0138 0.0126 ENR 7.3414 3.0600 3.3236 2.7834 2.0811 1.8931 2.0253 2.2346 EPI 9.2723 2.0816 2.2197 7.3521 3.2060 3.745 3.4477 4.5280

[0156] As shown in Table 4, the data comparison shows an improvement of 3.19 in PSNR, 0.1288 in SSIM, 0.0049 in MSE, 4.2814 in ENR, and 1.9202 in EPI. When processing Gaussian noise, the GBA algorithm outperforms other algorithms in all metrics, with excellent performance in overall detail and edge preservation.

[0157] The GBA algorithm demonstrates significant performance improvements by combining 3D chaotic mapping with the dung beetle optimization algorithm. The search space is further expanded under 3D chaotic mapping, and parameters are iteratively optimized to effectively preserve features during the filtering process. Comparing the noise reduction performance, the GBA algorithm is more effective than other existing filtering algorithms in preserving specific features. The GBA algorithm improves medical image quality by reducing noise, artifacts, and other defects, effectively assisting medical personnel in analyzing lesions and making more accurate judgments. The 3D chaotic mapping algorithm effectively ensures population uniformity and reduces convergence time when initializing the dung beetle population, thereby reducing the computational burden, accelerating processing time, and helping to reduce data redundancy, thereby simplifying the storage and transmission of medical images.

[0158] The above is only an embodiment of the present invention, and the common knowledge such as the specific technical solutions and / or characteristics in the solution are not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the description can be used to interpret the content of the claims.

Claims

1. A medical image denoising method based on 3D chaotic mapping dung beetle algorithm, characterized in that: The following steps are involved: Step 1: inputting an original medical image to be processed, performing noise processing on the original medical image, and generating a noisy image; Step 2: input the noisy image into a hybrid filter for denoising the image; Step 3, using the dung beetle optimization algorithm to calculate the optimal parameter combination of each filter in the hybrid filter, and using the optimal parameter combination to configure the hybrid filter; Step 4: De-noising the noisy image using the configured hybrid filter to generate a de-noised image; Among them, in step 3, the distribution position of the population is initialized using 3D chaotic mapping, and the 3D chaotic mapping is: Among them, x, y, z are system state variables, a=-0.6, c=-10, 1≤b≤7, and the initial values ​​are set to (0.1, 0, 0).

2. The medical image denoising method based on the 3D chaotic mapping dung beetle algorithm according to claim 1, characterized in that: In step 3, the optimal value of b is 3.

7.

3. The medical image denoising method based on the 3D chaotic mapping dung beetle algorithm according to claim 2, characterized in that: In step 2, the hybrid filter includes: a Gaussian filter, a median filter, a mean filter, a non-local mean filter and a sharpening filter.

4. The medical image denoising method based on the 3D chaotic mapping dung beetle algorithm according to claim 3, characterized in that: The dung beetle algorithm includes a rolling ball behavior. During the rolling behavior, the calculation formula for updating the dung beetle position is: x i (t+1) is the position of the i-th dung beetle at the t-th iteration, k∈(0,0.2] is a constant value indicating the deflection coefficient, b∈(0,1) is a constant; α is -1 or 1, X w is the global worst position, Δx is the degree of change in light intensity; When α=1, it indicates that the natural environment does not affect the original direction, and when α=-1, it indicates that the original direction is deviated from.

5. The medical image denoising method based on 3D chaotic mapping dung beetle algorithm according to claim 4, characterized in that: The dung beetle algorithm includes dancing behavior. During the dancing behavior, the calculation formula for the dung beetle's position update is: x i (t+1)=x i (t)+tan(θ)∣x i (t)-x i (t-1)∣? ∣x i (t)-x i (t-1)| is the difference between the position of the i-th dung beetle at the t-th iteration and the position at the t-1-th iteration; θ∈[0,π] is the deflection angle; when θ is equal to 0, π / 2, π, the dung beetle will not change its position.

6. The medical image denoising method based on 3D chaotic mapping dung beetle algorithm according to claim 5, characterized in that: The dung beetle algorithm includes a breeding behavior. During the breeding behavior, the calculation formula for the egg-laying area is: Lb * =max(X * ×(1-R),Lb),Ub * =min(X * ×(1+R),Ub); Lb * , Ub * are the lower and upper limits of the spawning area, Lb and Ub are the lower and upper limits of the search space, X * is the optimal position in the current population, and the inertia weight R = 1-t / T max , T max The maximum number of iterations during algorithm iteration; The calculation formula for the position of the incubation fecal ball is: B i (t+1)=X * +b1×(B i (t)-Lb * )+b2×(B i (t)-Ub * ); B i (t) is the position of the i-th incubated fecal ball at the t-th iteration, b1 and b2 represent two independent random vectors of size 1×D, and D is the dimension of the optimization problem.

7. The medical image denoising method based on 3D chaotic mapping dung beetle algorithm according to claim 6, characterized in that: The dung beetle algorithm includes foraging behavior. In the foraging behavior, the calculation formula of the optimal foraging area is: Lb b =max(X b ×(1-R),Lb),Ub b =min(X b ×(1+R),Ub); X b is the global optimal position, Lb b 、Ub b are the lower and upper limits of the optimal foraging area; The calculation formula for the position update of the dung beetle is: x i (t+1)=x i (t)+C1×(x i (t)-Lb b )+C2×(x i (t)-Ub b ); x i (t) represents the position information of the i-th dung beetle at the t-th iteration, C1 represents a random number following the normal distribution, and C2 represents a random vector belonging to (0,1).

8. The medical image denoising method based on 3D chaotic mapping dung beetle algorithm according to claim 7, characterized in that: The dung beetle algorithm includes stealing behavior. During the stealing behavior, the position update calculation formula of the thief dung beetle is: x i (t+1)=X b +S×g×(|x i (t)-X * |+|x i (t)-X b |); where x i (t) represents the position of the i-th thief in the t-th iteration, g is a random vector of size 1×D, which obeys the normal distribution, S represents a constant, X b To compete for the best position for food, X * is the optimal position in the current population.

Images