A fully automatic ultrasound liver fat content detection method and system based on double nakagami

Abstract

The present invention discloses a fully automatic ultrasound liver fat content detection method and system based on double Nakagami, comprising: S1. acquiring B-mode images of an ultrasound liver and IQ envelope data of an original RF signal; S2. performing image segmentation on the B-mode image using a deep learning segmentation method to segment out the liver parenchyma and vascular regions in the image; S3. post-processing the image segmentation results to obtain a candidate curved-edge trapezoidal ROI region containing only the liver parenchyma; S4. inputting the IQ envelope signal position area data corresponding to the candidate ROI region into a double Nakagami mixed distribution model, and estimating and optimizing the model parameters through an algorithm to obtain the fat content in the specified ROI region; the present invention realizes fully automatic quantitative detection of ultrasound liver fat content, which helps doctors or users complete the detection of liver fat content without relying on ultrasound clinical knowledge.

Description

Technical Field

[0001] The present invention relates to the technical field of liver fat measurement, and more particularly to a double-nakagami-based fully automatic ultrasonic liver fat content detection method and system. Background Art

[0002] Currently, ultrasound imaging is widely used in a variety of clinical settings to image different areas of the human body. Compared with other medical imaging modalities (such as magnetic resonance imaging (MRI) and computed tomography (CT), ultrasound imaging has many advantages, such as low cost, patient safety, easy access, high diagnostic efficiency, ease of use, portability, and no radiation.

[0003] A key application of ultrasound imaging is to help diagnose liver diseases, such as fatty liver and liver fibrosis. Ultrasound attenuation technology can achieve qualitative detection of liver fat content. In addition, the 2D-SWE and ARFI modes can measure liver hardness to determine the severity of liver fibrosis.

[0004] Although acoustic attenuation can achieve qualitative detection of liver fat content, that is, screening for mild, moderate and severe liver diseases, acoustic attenuation technology cannot achieve the purpose of quantitative detection and requires rich experience of ultrasound clinicians to make judgments. For example, the definition of mild fatty liver is 5% to 33% of liver fat content. The test result of mild can be used to judge that the fat content falls within this range, but the specific value is unknown. In addition, the range of 5% to 33% is also relatively large, so simply diagnosing mild fatty liver is not enough. If the fat content in the liver can be accurately quantified, it will be more meaningful to patients.

[0005] In addition, although ultrasound has many excellent features in liver diagnosis, clinicians and surgeons still face some challenges when using ultrasound imaging. For example, the ultrasound liver area includes blood vessels and speckle noise in addition to the liver parenchyma. If the imaging position is not accurate enough, ribs and other tissues may also be present. These tissues and areas other than the liver parenchyma will interfere with the detection of fatty liver and liver fibrosis.

[0006] Therefore, how to use ultrasound imaging to achieve fully automatic quantitative detection of liver fat is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a fully automatic ultrasonic liver fat content detection method and system based on double nakagami to solve some of the technical problems mentioned in the background technology.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A fully automatic ultrasound liver fat content detection method based on double nakagami comprises the following steps:

[0010] S1. Acquire B-mode ultrasound liver images and IQ envelope data of the original RF signal;

[0011] S2. Segment the B-mode image using a deep learning segmentation method to separate the liver parenchyma and vascular regions in the image;

[0012] S3. Post-processing the image segmentation results to obtain a candidate curved-edge trapezoidal ROI region containing only the liver parenchyma;

[0013] S4. Input the IQ envelope signal position area data corresponding to the candidate ROI area into the double-nakagami mixture distribution model to estimate and optimize the model parameters through the algorithm to obtain the fat content in the specified ROI area.

[0014] Preferably, the deep learning segmentation method in step S2 includes selecting a semantic segmentation network model or an instance segmentation network model to perform image segmentation on the ultrasound image of the liver.

[0015] Preferably, the specific contents of step S3 include:

[0016] S31. Acquire an image after segmentation of the liver parenchyma and vascular regions;

[0017] S32. Combined with the center coordinates, large circle radius, small circle radius and center angle of the ultrasound system settings, the input image is divided into several curved trapezoidal ROI areas;

[0018] S33. Determine whether each ROI region is liver parenchyma based on the segmentation results of the liver parenchyma and vascular regions, and retain all ROI regions that meet the conditions as candidate ROI regions;

[0019] S34. Perform inverse transformation of scan conversion on the candidate ROI regions to obtain the rectangular region of the IQ envelope signal corresponding to each ROI region.

[0020] Preferably, the double-Nakagami mixed distribution model includes two Nakagami distributions with different weights, and the probability density formula of the model is:

[0021] p d =(1-α)p L (x|m L ,Ω L )+αp F (x|m F ,Ω F )

[0022] Among them, p L(x|m L ,Ω L ) is the Nakagami distribution obtained from liver tissue, p F (x|m F ,Ω F ) is the Nakagami distribution obtained from lipid droplets, m L Characterizes the tissue characteristics of the liver with the liver parenchyma and luminal structures as the main scattering sources, m F Characterize the tissue characteristics of fat droplets as the scattering source, x is the signal amplitude of the ultrasound echo, that is, the data in the corresponding ROI area of the IQ envelope signal, and α is the sample weight of the fat droplet distribution, that is, the fat content in the specified ROI area;

[0023] The probability density function of the Nakagami distribution is:

[0024]

[0025] where Γ(·) is the gamma function, U(x) is the step function, m is the shape parameter of nakagami, and Ω is the scale parameter.

[0026] Preferably, the parameter estimation of the double-nakagami mixture distribution model includes: pre-processing the data in the IQ envelope signal ROI and inputting it into the double-nakagami mixture distribution model, optimizing the mixture distribution model by using the KL divergence optimization method or the EM expectation maximization method until the preset convergence condition is reached, and estimating the shape parameter m L and m F , scale parameter Ω L and Ω F , and weight α, α is the fat content in the specified ROI area.

[0027] Preferably, the method for calculating the parameters of the hybrid model using the EM algorithm is specifically as follows:

[0028] Initialize the model parameters, including the weights α and 1-α of the two nakagami distributions, and the shape parameter m L and m F , and the scale parameter Ω L and Ω F ;

[0029] First, the initialized model parameters are used to calculate the probability of each sample point in the sample space belonging to each nakagami distribution through E steps;

[0030] Calculate the parameters of the hybrid model through the M step according to the calculation results of the E step;

[0031] Set the specified convergence conditions, including the absolute value difference of the parameters estimated in the two M steps or the maximum number of iterations, and continuously iterate the E step and M step. When the convergence conditions are met, the last parameter value is the estimated parameter value of the double-nakagami mixture distribution model.

[0032] Preferably, the method for calculating the parameters of the hybrid model using the KL algorithm is as follows:

[0033] After preprocessing, the IQ envelope signal is normalized and the parameters of the double-nakagami mixture distribution model are initialized;

[0034] Calculate the cumulative probability density function of the double-Nakagami mixture distribution model based on the initialization parameters, calculate the horizontal coordinate value when the cumulative probability density function is close to 1, and multiply it by the normalized data range of the IQ envelope signal. Then calculate its histogram as the probability density value, and then divide the horizontal coordinate value by the histogram probability density value and calculate the KL divergence with the theoretical probability density of the double-Nakagami mixture distribution model.

[0035] Select the Nelder-Mead optimization method to optimize the KL divergence. By setting the Nelder-Mead termination condition, including the maximum number of iterations or the difference between the two estimation results, when the specified termination condition is reached, the current value of the calculated KL divergence is the parameter value of the double-nakagami mixture distribution model.

[0036] A fully automatic ultrasound liver fat content detection system based on double nakagami, based on the described fully automatic ultrasound liver fat content detection method based on double nakagami, comprising: an ultrasound detection device, an image segmentation module, a post-processing module and a DND module;

[0037] Ultrasound detection equipment for obtaining ultrasound images of the liver;

[0038] An image segmentation module is used to segment the ultrasound image of the liver using a deep learning segmentation method to separate the liver parenchyma and vascular areas in the image;

[0039] A post-processing module is used to process the image segmentation results to obtain a curved trapezoidal ROI region containing only the liver parenchyma;

[0040] The DND module is used to estimate and optimize the model parameters of the post-processed ROI area through the double-Nakagami mixed distribution model algorithm to obtain the fat content in the specified ROI area.

[0041] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a fully automatic ultrasonic liver fat content detection method based on double nakagami.

[0042] A processing terminal includes a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor executes the computer program, the fully automatic ultrasonic liver fat content detection method based on double nakagami is implemented.

[0043] As can be seen from the above technical solution, compared with the existing technology, the present invention discloses a fully automatic ultrasound liver fat content detection method and system based on double Nakagami. Based on the IQ envelope data of the original ultrasound RF radio frequency signal, this data has not undergone post-processing operations such as image enhancement, and its statistical characteristics include tissue structure information. The double Nakagami mixed distribution model can accurately model the liver parenchyma and fat droplets in the IQ envelope data separately. By calculating the weight of each component, the liver fat content is obtained, realizing fully automatic quantitative detection of ultrasound liver fat content.

[0044] Compared with liver puncture to measure fat content, the present invention relies on the statistical characteristics of ultrasonic echoes. It is a non-invasive technology that does not require skin penetration, reducing the risk of infection. At the same time, it makes up for the disadvantage that acoustic attenuation can only detect fatty liver qualitatively but not quantitatively. The automatic segmentation and post-processing scheme adopted helps doctors or users complete the detection of liver fat content without relying on ultrasonic clinical knowledge. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0046] Figure 1 A schematic diagram of a fully automatic ultrasound liver fat content detection method based on double nakagami provided by the present invention;

[0047] Figure 2 Schematic diagram of the liver and blood vessel segmentation results using Yolov8 provided by the present invention;

[0048] Figure 3 A schematic diagram of the post-processing method provided by the present invention;

[0049] Figure 4A schematic diagram of dividing the curved-edge trapezoidal ROI area provided by the present invention;

[0050] Figure 5 A schematic diagram of a candidate ROI provided by the present invention that only includes the liver parenchyma area;

[0051] Figure 6 Schematic diagram of the DND model parameter estimation method provided by the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] The embodiment of the present invention discloses a fully automatic ultrasonic liver fat content detection method based on double nakagami, such as Figure 1 , including the following steps:

[0054] S1. Acquire B-mode ultrasound liver images and IQ envelope data of the original RF signal;

[0055] S2. Segment the B-mode image using a deep learning segmentation method to separate the liver parenchyma and vascular regions in the image;

[0056] S3. Post-processing the image segmentation results to obtain a candidate curved-edge trapezoidal ROI region containing only the liver parenchyma;

[0057] S4. Inputting the IQ envelope signal position area data corresponding to the candidate ROI region into a double nakagami mixture distribution model (DND model) and using an algorithm to estimate and optimize the model parameters to obtain the fat content in the specified ROI region.

[0058] In order to further implement the above technical solution, the deep learning segmentation method in step S2 includes selecting a semantic segmentation network model or an instance segmentation network model to perform image segmentation on the ultrasound image of the liver.

[0059] In this embodiment, the data set of the deep learning segmentation method is collected by recruiting multiple patients. For example, 50 patients can be recruited for collection, and the amount of data collected is 2000. Then the ratio of the training set, the validation set and the test set is divided according to 8:1:1. Semantic segmentation network models such as u2net and unet++ or instance segmentation network models such as maskrcnn, yolact and yolov8 are selected for training. For example, the segmentation result of the ultrasound liver image in the yolov8 instance segmentation network is as follows: Figure 2 As can be seen from the figure, Yolov8 can effectively distinguish blood vessels and liver parenchyma from other tissues and background, thus achieving segmentation. This is also possible even when the blood vessels are relatively small. After actual testing, the Yolov8 model achieved an accuracy rate of 98% for liver parenchyma segmentation.

[0060] In order to further implement the above technical solutions, Figure 3 , the specific contents of step S3 include:

[0061] S31. Acquire an image after segmentation of the liver parenchyma and vascular regions;

[0062] S32. Combine the center coordinates, large circle radius, small circle radius and center angle of the ultrasound system settings to divide the input image into several curved trapezoidal ROI areas. After visualization, Figure 4 ;

[0063] S33. Combine the segmentation results of liver parenchyma and vascular regions to determine whether each ROI region is liver parenchyma, and retain all ROI regions that meet the conditions, namely candidate ROI regions, such as Figure 5 ;

[0064] S34. Perform inverse transformation of scan conversion on the candidate ROI regions to obtain the rectangular regions of the IQ envelope signals corresponding to the respective ROI regions.

[0065] To further implement the above technical solution, the double-Nakagami mixed distribution model contains two Nakagami distributions with different weights. The probability density formula of the model is:

[0066] p d =(1-α)p L (x|m L ,Ω L )+αp F (x|m F ,Ω F )

[0067] Among them, p L (x|m L ,Ω L) is the Nakagami distribution obtained from liver tissue, p F (x|m F ,Ω F ) is the Nakagami distribution obtained from lipid droplets, m L Characterizes the tissue characteristics of the liver with the liver parenchyma and luminal structures as the main scattering sources, m F Characterizes the tissue characteristics of fat droplets as the scattering source, x is the signal amplitude of the ultrasound echo, that is, the data in the corresponding ROI area of the IQ envelope signal, and α represents the fat content in the specified ROI area;

[0068] The probability density function of the Nakagami distribution is:

[0069]

[0070] Where Γ(·) is the gamma function, U(x) is the step function, m is the shape parameter of Nakagami, and Ω is the scale parameter. Different tissues are mainly characterized by differences in the shape parameter m. For example, the shape parameter m of liver parenchyma is less than 0.8, while the shape parameter of lipid droplets is generally greater than 0.9.

[0071] In order to further implement the above technical solutions, Figure 6 The parameter estimation of the double-nakagami mixture distribution model is as follows: the data in the IQ envelope signal ROI is preprocessed and input into the double-nakagami mixture distribution model, and the mixture distribution model is optimized by the Kullback-Leibler divergence optimization method (referred to as KL algorithm) or the expectation maximization algorithm (referred to as EM algorithm) until the preset convergence condition is reached, and the shape parameter m is estimated. L and m F , scale parameter Ω L and Ω F , and weight Ω, α is the fat content in the specified ROI area.

[0072] In this embodiment, the purpose of preprocessing is to remove the interference of noise. The preprocessing method is to first arrange the input data from small to large, then discard the front and back 5% of the data to eliminate the influence of too large and too small data, and at the same time select a suitable window size for Gaussian smoothing of the remaining data.

[0073] In order to further implement the above technical solution, the method of calculating the parameters of the hybrid model using the EM algorithm is as follows:

[0074] Initialize the model parameters, including the weights α and 1-α of the two nakagami distributions, and the shape parameter m L and mF , and the scale parameter Ω L and Ω F ;

[0075] In this embodiment, weight initialization is achieved using k_means clustering. After instance segmentation, the input data consists of liver parenchyma and fat regions, excluding blood vessels and other tissue regions. Therefore, the number of K_means clusters is set to 2. The results obtained are the proportions of liver data and fat data respectively.

[0076] E-step: The first step uses the initialized model parameters to calculate the probability of each sample point in the sample space belonging to each nakagami distribution through E-step; starting from the second step, the parameters of each model are calculated using the results of M-step;

[0077] In this embodiment, the calculation method of step E is:

[0078]

[0079] Where j is the number of each individual nakagami distribution in the double nakagami distribution, i is the number of the input data, and the result is the probability that each sample belongs to each nakagami distribution;

[0080] M-step: Calculate the parameters of the hybrid model through the M-step according to the calculation results of the E-step;

[0081] The specific calculation formula is:

[0082]

[0083] Set the specified convergence conditions, including the absolute value difference of the parameters estimated in the two M steps or the maximum number of iterations, and continuously iterate the E step and M step. When the convergence conditions are reached, the last parameter value is the estimated DND mixed distribution model parameter value, including α, m L 、m F ,Ω L ,Ω F .

[0084] In order to further implement the above technical solution, the method of calculating the parameters of the hybrid model using the Kullback-Leibler divergence optimization method is as follows:

[0085] After preprocessing, the IQ envelope signal is normalized and the DND model parameters are initialized. The normalization method is to divide it by the maximum value of the group of data;

[0086] The divergence that the KL algorithm wants to optimize is:

[0087]

[0088] Among them, q(x) and p(x) are the probability distribution of the observed IQ envelope signal and the distribution of the theoretically fitted DND mixed distribution model, respectively. The probability distribution of q(x) is approximated by the histogram probability density of the IQ envelope signal. Because the numerical range of the input IQ data has been normalized, but the maximum value of the numerical range of the DND model is theoretically greater than 1, under different shape parameters and scale parameters, the maximum value of the probability density of the DND model is greater than 5. Therefore, the probability density curve of the DND model needs to be processed before the KL divergence formula can be applied for optimization.

[0089] The processing method is as follows: calculate the cumulative probability density function (CDF) of the DND model according to the initialization parameters, then calculate the horizontal coordinate value γ when the cumulative probability density function is close to 1, multiply γ by the normalized data range of q(x), and then calculate the probability density of the histogram of q(x). Finally, in the process of calculating the KL divergence, divide γ by the histogram probability density value of q(x) and then calculate the KL divergence with the theoretically fitted DND probability density p(x);

[0090] Select the Nelder-Mead optimization method to optimize the KL divergence. By setting the Nelder-Mead termination condition, including the maximum number of iterations or the difference between the two estimation results, when the specified termination condition is reached, the current value of the calculated KL divergence is the parameter value of the double-nakagami mixture distribution model.

[0091] A fully automatic ultrasound liver fat content detection system based on double nakagami is based on a fully automatic ultrasound liver fat content detection method based on double nakagami, comprising: an ultrasound detection device, an image segmentation module, a post-processing module and a DND module;

[0092] Ultrasound detection equipment for obtaining ultrasound images of the liver;

[0093] An image segmentation module is used to segment the ultrasound image of the liver using a deep learning segmentation method to separate the liver parenchyma and vascular areas in the image;

[0094] A post-processing module is used to process the image segmentation results to obtain a curved trapezoidal ROI region containing only the liver parenchyma;

[0095] The DND module is used to estimate and optimize the model parameters of the post-processed ROI area through the double-Nakagami mixed distribution model algorithm to obtain the fat content in the specified ROI area.

[0096] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a fully automatic ultrasound liver fat content detection method based on double nakagami.

[0097] A processing terminal includes a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor executes the computer program, a full-automatic ultrasonic liver fat content detection method based on double nakagami is implemented.

[0098] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0099] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fully automatic ultrasound liver fat content detection method based on double nakagami, characterized in that: The following steps are involved: S1. Acquire B-mode ultrasound liver images and IQ envelope data of the original RF signal; S2. Segment the B-mode image using a deep learning segmentation method to separate the liver parenchyma and vascular regions in the image; S3. Post-processing the image segmentation results to obtain a candidate curved-edge trapezoidal ROI region containing only the liver parenchyma; S4. Inputting the IQ envelope signal position area data corresponding to the candidate ROI region into the double-nakagami mixture distribution model to estimate and optimize the model parameters through the algorithm to obtain the fat content in the specified ROI region; The specific contents of step S3 include: S31. Acquire an image after segmentation of the liver parenchyma and vascular regions; S32. Combined with the center coordinates, large circle radius, small circle radius and center angle of the ultrasound system settings, the input image is divided into several curved trapezoidal ROI areas; S33. Determine whether each ROI region is liver parenchyma based on the segmentation results of the liver parenchyma and vascular regions, and retain all ROI regions that meet the conditions as candidate ROI regions; S34. Perform an inverse transform of the scan conversion on the candidate ROI region to obtain a rectangular region of the IQ envelope signal corresponding to each ROI region; The KL algorithm is used to calculate the parameters of the mixed distribution model, specifically: After preprocessing, the IQ envelope signal is normalized and the parameters of the double-nakagami mixture distribution model are initialized; Calculate the cumulative probability density function of the double-Nakagami mixture distribution model based on the initialization parameters, calculate the horizontal coordinate value when the cumulative probability density function is close to 1, and multiply it by the normalized data range of the IQ envelope signal. Then calculate its histogram as the probability density value, and then divide the horizontal coordinate value by the histogram probability density value and calculate the KL divergence with the theoretical probability density of the double-Nakagami mixture distribution model. Select the Nelder-Mead optimization method to optimize the KL divergence. By setting the Nelder-Mead termination condition, including the maximum number of iterations or the difference between the two estimation results, when the specified termination condition is reached, the current value of the calculated KL divergence is the parameter value of the double-nakagami mixture distribution model.

2. The method for detecting liver fat content by double-nakagami-based full-automatic ultrasound according to claim 1, characterized in that: The deep learning segmentation method in step S2 includes selecting a semantic segmentation network model or an instance segmentation network model to perform image segmentation on the ultrasound image of the liver.

3. The method for detecting liver fat content by double-nakagami-based full-automatic ultrasound according to claim 1, characterized in that: The double-Nakagami mixture distribution model contains two Nakagami distributions with different weights. The probability density formula of the model is: p d =(1-α)p L (x|m L ,Oh L )+αp F (x|m F ,Oh F ) Among them, p L (x|m L ,Ω L ) is the Nakagami distribution obtained from liver tissue, p F (x|m F ,Ω F ) is the Nakagami distribution obtained from lipid droplets, m L Characterizes the tissue characteristics of the liver with the liver parenchyma and luminal structures as the main scattering sources, m F Characterize the tissue characteristics of fat droplets as the scattering source, x is the signal amplitude of the ultrasound echo, that is, the data in the corresponding ROI area of the IQ envelope signal, and α is the sample weight of the fat droplet distribution, that is, the fat content in the specified ROI area; The probability density function of the Nakagami distribution is: where Γ(·) is the gamma function, U(x) is the step function, m is the shape parameter of nakagami, and Ω is the scale parameter.

4. The method for detecting liver fat content by double-nakagami-based automatic ultrasound according to claim 3, characterized in that: The parameter estimation of the double-Nakagami mixture distribution model is as follows: the data in the IQ envelope signal ROI is preprocessed and input into the double-Nakagami mixture distribution model, the KL divergence optimization method or the EM expectation maximization method is used to optimize the mixture distribution model until the preset convergence condition is reached, and the shape parameter m is estimated. L and m F , scale parameter Ω L and Ω F , and weight α, α is the fat content in the specified ROI area.

5. The method for detecting liver fat content by double-nakagami-based full-automatic ultrasound according to claim 4, characterized in that: The specific method of calculating the parameters of the mixed distribution model using the EM algorithm is: Initialize the model parameters, including the weights α and 1-α of the two nakagami distributions, and the shape parameter m L and m F , and the scale parameter Ω L and Ω F ; First, the initialized model parameters are used to calculate the probability of each sample point in the sample space belonging to each nakagami distribution through E steps; Calculate the parameters of the mixed distribution model through the M step according to the calculation results of the E step; Set the specified convergence conditions, including the absolute value difference of the parameters estimated in the two M steps or the maximum number of iterations, and continuously iterate the E step and M step. When the convergence conditions are met, the last parameter value is the estimated parameter value of the double-nakagami mixture distribution model.

6. A fully automatic ultrasound liver fat content detection system based on double nakagami, characterized in that: A fully automatic ultrasound liver fat content detection method based on double nakagami according to any one of claims 1 to 5, comprising: an ultrasound detection device, an image segmentation module, a post-processing module and a DND module; Ultrasound detection equipment for obtaining ultrasound images of the liver; An image segmentation module is used to segment the ultrasound image of the liver using a deep learning segmentation method to separate the liver parenchyma and vascular areas in the image; A post-processing module is used to process the image segmentation results to obtain a curved trapezoidal ROI region containing only the liver parenchyma; The DND module is used to estimate and optimize the model parameters of the post-processed ROI area through the double-Nakagami mixed distribution model algorithm to obtain the fat content in the specified ROI area.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program implements the method for fully automatic liver fat content detection based on double nakagami according to any one of claims 1 to 5.

8. A processing terminal comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, the fully automatic ultrasonic liver fat content detection method based on double nakagami as described in any one of claims 1 to 5 is implemented.

Images