A multi-frame image fat liver content detection method and system based on motion suppression

Abstract

The application discloses a kind of multi-frame image fatty liver content detection method and system based on motion inhibition, belong to medical image processing technical field, which comprises: selecting region of interest in patient liver region and setting processing time;Under B mode ultrasonic scanning, liver image is acquired in real time;Liver parenchyma probability weight distribution map is generated to each frame image;Interframe motion estimation is carried out, and the image frame with motion amplitude in preset threshold is screened out;Nonlinear parameter scanning is carried out, and corresponding nonlinear parameter matrix is obtained;Each element in each nonlinear parameter matrix is converted into liver fat content value, and corresponding liver fat content score matrix is obtained;Kalman filtering algorithm is applied to multiple liver fat content score matrixes to carry out fusion filtering processing, and finally liver fat content quantitative value is obtained.The method can realize non-invasive, accurate and quantitative liver fat content detection;It can be integrated into ultrasonic diagnostic equipment and is suitable for hospitals, medical centers and other scenes.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and more specifically to a method and system for detecting fatty liver content in multi-frame images based on motion inhibition. Background Technology

[0002] Metabolic associated fatty liver disease (MAFLD) is a liver disease closely related to insulin resistance and metabolic dysfunction, encompassing a spectrum from simple steatosis (MAFLD) to metabolic associated steatohepatitis (MAASH). With the global prevalence of obesity and metabolic syndrome, MAFLD has become a leading cause of chronic liver disease, and some patients may progress to liver fibrosis, cirrhosis, or even liver cancer. Therefore, early, non-invasive, and accurate quantitative detection of liver fat content has significant clinical value.

[0003] Currently, the diagnosis and assessment of liver fat content mainly rely on two methods: liver biopsy and imaging examinations. Liver biopsy is considered the gold standard for detecting liver fat content, directly assessing the degree of steatosis, inflammation, and fibrosis through histological examination. However, it has problems such as invasiveness, sampling errors, and the risk of complications, and is difficult to use for dynamic monitoring. Imaging examinations include conventional ultrasound, CT / MRI, and CAP (controlled attenuation parameter) examination. Conventional ultrasound qualitatively determines the severity of fatty liver by the intensity of liver echoes, but it cannot be quantitative and is greatly affected by the operator's experience. MRI's PDFF (proton density fat fraction) is currently the most accurate quantitative imaging method, but its cost per test is high, it is highly dependent on equipment, and it is difficult to popularize. CT can also qualitatively detect fat content (such as the CT value ratio of liver to spleen), but CT involves radiation and cannot be used as a long-term screening method. Finally, CAP technology based on ultrasound platforms can qualitatively detect the mild, moderate, and severe grades of fatty liver, but it still cannot perform quantitative detection.

[0004] The relationship between the sound pressure and the density of the medium during ultrasound propagation in biological tissues can be approximated using the Taylor technique:

[0005]

[0006] In the formula, p represents sound pressure, which is the pressure change generated when ultrasound propagates in a medium.

[0007] It represents the change in the density of the medium, that is, the density fluctuation caused by sound wave disturbance.

[0008] It represents the initial density of the medium, that is, the original density of the medium before the sound wave propagates.

[0009] A represents the coefficient of the first term, reflecting the linear relationship between sound pressure and density change, and is related to the linear acoustic properties of the medium.

[0010] B represents the quadratic coefficient, characterizing the intensity of nonlinear effects during sound propagation. Its ratio to the linear coefficient A, B / A, is defined as the nonlinear parameter of the medium and is a key indicator for quantifying the nonlinear characteristics of an organism.

[0011] C represents the coefficient of the cubic term, corresponding to higher-order nonlinear effects. In practical applications, since the influence of cubic and higher-order terms is usually small, they are often ignored to simplify the model.

[0012] L represents terms of fourth order or higher.

[0013] In the above formula, the ratio B / A of the coefficients of the quadratic term and the linear term is defined as the nonlinear parameter of the medium.

[0014] Table 1 below shows the magnitude of nonlinear parameters for different tissue types in the human body. As can be seen from Table 1, there is a significant difference between the nonlinear parameter values ​​of a healthy liver and those of a liver in a diseased state. In particular, the nonlinear parameter values ​​of a healthy liver and fat differ greatly, being 6.8±0.5 and 9±0.5, respectively.

[0015] Table 1 Nonlinear parameter values ​​for different human tissues

[0016]

[0017] The classic method for measuring nonlinear parameters is to use thermodynamics, which estimates the magnitude of the nonlinear parameter by utilizing the change in wave velocity with sound pressure under isothermal conditions. Although this method is relatively accurate, the experimental setup and method are very complex, making it unsuitable as a diagnostic method in clinical settings.

[0018] Patent document CN104757999A discloses a nonlinear imaging method based on ultrasound fundamental wave and harmonics. However, the image generated by this method does not directly participate in the nonlinear calculation, but rather generates an image that is correlated with the nonlinear parameters. Patent CN107970042A uses fundamental wave and harmonics under high and low voltage to generate nonlinear parameter images. Imaging a specified ROI region of the liver with this technology will generate a nonlinear parameter pseudo-color image. The darker the color, the larger the nonlinear parameter, which represents a high liver fat content. Conversely, the smaller the nonlinear parameter, which represents a low liver fat content. Therefore, this method can only obtain a qualitative result of liver fat content.

[0019] Therefore, the method of generating nonlinear parameter images based on fundamental and harmonic waves in related technologies can only achieve qualitative analysis and cannot quantitatively output fat content values. Furthermore, it is susceptible to interference from motion, artifacts, noise, and non-liver tissues. Summary of the Invention

[0020] This invention aims to solve the problem that existing technologies cannot achieve non-invasive, accurate, and quantitative detection of liver fat content, and proposes a method and system for detecting fatty liver content based on motion inhibition multi-frame images.

[0021] To achieve the above objectives, the present invention adopts the following technical solution:

[0022] In a first aspect, embodiments of the present invention provide a method for detecting fatty liver content in multi-frame images based on motion inhibition, comprising the following steps:

[0023] Select the region of interest in the patient's liver area and set the processing time; acquire liver images in real time under B-mode ultrasound scanning;

[0024] Generate a probability weight distribution map of liver parenchyma for each frame of image;

[0025] Perform inter-frame motion estimation and filter out image frames whose motion amplitude is within a preset threshold;

[0026] The selected image frames are scanned using nonlinear parameters to obtain the corresponding nonlinear parameter matrix;

[0027] Each nonlinear parameter value in each of the aforementioned nonlinear parameter matrices is converted into a liver fat content score value to obtain the corresponding liver fat content score matrix.

[0028] The Kalman filter algorithm is applied to fuse and filter multiple liver fat content fraction matrices to obtain the final quantitative value of liver fat content.

[0029] Furthermore, the probability weight distribution map of the liver parenchyma is generated by a deep learning semantic segmentation model; the model adopts a U-Net network structure, and during training, only pixels in the liver parenchyma region are labeled, while non-liver parenchyma regions include blood vessels, shadows, calcifications and other non-liver tissues; the output is the probability value of each pixel belonging to the liver parenchyma.

[0030] Further, inter-frame motion estimation is performed to select image frames whose motion amplitude is within a preset threshold; including:

[0031] The first frame captured in the image sequence is set as the reference frame, and the frames captured after it are set as the current frames in sequence.

[0032] The regions of interest in the reference frame and the current frame are respectively divided into multiple image sub-blocks of size M×N pixels;

[0033] In the current frame, for each target image sub-block in the reference frame, block matching calculation is performed in the corresponding search area to find the optimal matching sub-block;

[0034] The relative motion amplitude between the current frame and the reference frame is calculated based on the positional difference between the target image sub-block and the optimal matching sub-block.

[0035] The motion amplitude is compared with a distance threshold. If the motion amplitude is less than or equal to the distance threshold, the current frame is retained for subsequent nonlinear parameter calculation; otherwise, the current frame is discarded.

[0036] Furthermore, the matching criteria used in the block matching calculation are the sum of absolute differences, squared errors, or normalized cross-correlation functions.

[0037] Furthermore, a nonlinear parameter scan is performed on the selected image frames to obtain the corresponding nonlinear parameter matrix; including:

[0038] Four sets of echo signals were acquired: high-voltage fundamental, high-voltage harmonic, low-voltage fundamental, and low-voltage harmonic.

[0039] Perform median filtering on the signal;

[0040] Construct and solve the nonlinear parametric equations to obtain the corresponding nonlinear parameter matrix.

[0041] Further, each nonlinear parameter value in each of the aforementioned nonlinear parameter matrices is converted into a liver fat content score, resulting in a corresponding liver fat content score matrix; including:

[0042] The nonlinear parameter values ​​for obtaining the ideal liver parenchyma are: Nonlinear parameter values ​​of fat :

[0043] Based on the linear relationship between the nonlinear parameter values ​​and fat content in a specified region of the liver, the following linear relationship model is constructed:

[0044]

[0045] f represents the fat content corresponding to each nonlinear parameter value. These are the nonlinear parameter values ​​corresponding to liver parenchyma without fat. P represents the non-linear parameter value corresponding to fat; P represents the non-linear parameter value corresponding to a certain pixel, i.e., the grayscale value.

[0046] Based on the linear relationship model, each element in each nonlinear parameter matrix is ​​converted into a liver fat content score between 0 and 1, resulting in a liver fat content score matrix.

[0047] Furthermore, the Kalman filter algorithm is applied to fuse and filter multiple liver fat content fraction matrices to obtain the final quantitative value of liver fat content; including:

[0048] Establish the state equation and the observation equation;

[0049] Initialize the state vector and error covariance matrix;

[0050] Perform prediction and update steps, and adjust the measurement noise covariance by incorporating liver parenchyma probability weights;

[0051] The mean of the optimal fat content score matrix is ​​output as the final liver fat content.

[0052] Secondly, embodiments of the present invention also provide a multi-frame image fatty liver content detection system based on motion inhibition, using the method described in any of the first aspects, the system comprising:

[0053] The ultrasound imaging module is used to select the region of interest in the patient's liver area and set the processing time; it acquires liver images in real time under B-mode ultrasound scanning.

[0054] The liver parenchyma probability weight calculation module is used to generate a liver parenchyma probability weight distribution map for each frame of image;

[0055] The motion estimation module is used to perform inter-frame motion estimation and filter out image frames whose motion amplitude is within a preset threshold.

[0056] The nonlinear parameter calculation module is used to perform nonlinear parameter scanning on the filtered image frames to obtain the corresponding nonlinear parameter matrix.

[0057] The fat content conversion module is used to convert each nonlinear parameter value in each of the nonlinear parameter matrices into a liver fat content score value, thereby obtaining the corresponding liver fat content score matrix.

[0058] The Kalman filter module applies the Kalman filter algorithm to perform fusion filtering on multiple liver fat content fraction matrices to obtain the final quantitative value of liver fat content.

[0059] As can be seen from the above technical solution, compared with the prior art, the present invention has the following technical effects:

[0060] This invention predicts liver fat content based on the joint analysis of multiple ultrasound images, utilizes the probability distribution of liver parenchyma to weaken interference from non-liver tissues, and introduces Kalman filtering to suppress noise and outliers. This invention can be integrated into ultrasound diagnostic equipment and is suitable for hospitals, health check-up centers, and other settings, providing a quantitative tool for fatty liver screening and follow-up, offering strong stability and more reliable results. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0062] Figure 1 The flowchart of the multi-frame image fatty liver content detection method based on motion inhibition provided by the present invention is shown.

[0063] Figure 2 This is a schematic diagram illustrating the principle of fatty liver content detection provided by the present invention.

[0064] Figure 3 This is a schematic diagram of the segmentation results of liver images provided by the present invention using the U_Net network model.

[0065] Figure 4 The flowchart for inter-frame motion estimation calculation provided by this invention is shown.

[0066] Figure 5 This is a schematic diagram of the ROI region of a liver image provided by the present invention.

[0067] Figure 6 The flowchart for obtaining the nonlinear parameter matrix provided by this invention is shown.

[0068] Figure 7 A schematic diagram of the nonlinear parameter A / B matrix provided by the present invention.

[0069] Figure 8 This is a schematic diagram of the liver fat content fraction matrix provided by the present invention.

[0070] Figure 9 This is a block diagram of a multi-frame image fatty liver content detection system based on motion inhibition provided by the present invention. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] This invention discloses a multi-frame image-based method for detecting fatty liver content based on motion suppression. It is a method for quantitatively measuring liver fat content based on the physical characteristic that healthy livers and fat have different nonlinear parameters. This method effectively shields the problem of excessive data fluctuations caused by respiration and probe jitter during liver scanning. Furthermore, considering that the nonlinear parameters are affected by factors such as blood vessels, non-liver tissues, and noise in addition to liver parenchyma and fat, the actual measurement results may differ from the theoretical values. In addition to these issues, patients cannot remain completely still during the detection process, or slight cardiac traction can lead to differences in measurement results between consecutive frames. Therefore, this invention proposes a multi-frame-based Kalman filter-based method for detecting liver fat content, which is more stable and reliable than previous techniques.

[0073] like Figure 1 As shown in the figure, this invention discloses a method for detecting fatty liver content in multi-frame images based on motion inhibition, including the following steps:

[0074] S1. Select the region of interest in the patient's liver area and set the processing time; acquire liver images in real time under B-mode ultrasound scanning;

[0075] S2. Generate a probability weight distribution map of liver parenchyma for each frame of image;

[0076] S3. Perform inter-frame motion estimation and filter out image frames whose motion amplitude is within a preset threshold.

[0077] S4. Perform nonlinear parameter scanning on the filtered image frames to obtain the corresponding nonlinear parameter matrix;

[0078] S5. Convert each nonlinear parameter value in each of the nonlinear parameter matrices into a liver fat content score value to obtain the corresponding liver fat content score matrix.

[0079] S6. Apply the Kalman filter algorithm to perform fusion filtering on the multiple liver fat content fraction matrices to obtain the final quantitative value of liver fat content.

[0080] This method enables non-invasive, accurate, and quantitative detection of liver fat content; robustness is improved through multi-frame fusion and motion estimation; measurement accuracy is enhanced through liver parenchyma probability weighting; and noise and outlier effects are suppressed using Kalman filtering.

[0081] The technical solution of the present invention will be described in detail below:

[0082] 1. Reference Figure 2The diagram shows the overall flowchart for target liver detection. First, a suitable Region of Interest (ROI) is selected at the location of the patient's liver, and a processing time is set. The purpose of the processing time is to set a certain time range to collect enough information. Then, the ultrasound B_mode is turned on so that subsequent steps can perform tissue probability weight calculation and frame-to-frame motion estimation.

[0083] 2. During B-mode scanning, a probability weight distribution map of liver parenchyma is generated for the current frame image. This can be achieved using a deep learning-based segmentation method or other traditional image processing algorithms. If using deep learning, a semantic segmentation model can be employed, such as the U-Net network model. During model training, the gold standard only labels the liver parenchyma, while other areas such as blood vessels, shadows, calcified regions, and other non-liver tissues are labeled as non-liver tissues. Thus, the output during the forward inference phase after model training is the probability distribution of liver parenchyma. Regions with high probability values ​​(e.g., greater than 0.9) are considered liver parenchyma; those with probabilities less than a specified threshold (e.g., less than 0.6) may be shadowed areas behind blood vessels; and those with probabilities around 0.1 are blood vessel regions or other parts that are clearly not liver parenchyma. The output probability map is retained for subsequent Kalman filtering.

[0084] Reference Figure 3 The image shown is the segmentation result of a liver image using the U_Net network model. Figure 3 The left half of the image is the original ultrasound B-mode image, and the right half is the segmentation result based on the U-Net network model. The green area represents the segmented liver parenchyma, and the red area represents the segmented blood vessels. The probability map threshold for the segmentation output is 0.5, meaning that areas with a probability greater than 0.5 in the output probability map are considered liver parenchyma, and areas with a probability less than 0.5 are considered background.

[0085] After obtaining the segmentation probability map, the segmentation probability map of the liver is retained, which is called the tissue probability weight map W. Each element of the weight map is between 0 and 1, where the closer to 1 is, the greater the probability of it being liver parenchyma, and vice versa. The significance of retaining this tissue probability weight is to filter out regions with low tissue probability weights in the subsequent Kalman filtering process.

[0086] 3. Motion estimation: This is used to determine the magnitude of motion between frames. Since fat in the liver is not uniformly distributed, if the motion amplitude of the region between consecutive frames is too large, the fat content inside may also be different and unpredictable. Therefore, this embodiment of the invention aims to avoid this situation and uses motion estimation to determine the magnitude of motion between frames.

[0087] Motion estimation includes the following steps 1) to 5):

[0088] 1) Set the first frame captured in the image sequence as the reference frame, and set the subsequent frames as the current frames in sequence;

[0089] 2) Divide the regions of interest in the reference frame and the current frame into multiple image sub-blocks of size M×N pixels;

[0090] 3) In the current frame, for the target image sub-block in the reference frame, perform block matching calculation within the corresponding search area to find the optimal matching sub-block;

[0091] 4) Calculate the relative motion amplitude between the current frame and the reference frame based on the positional difference between the target image sub-block and the optimal matching sub-block;

[0092] 5) Compare the motion amplitude with the distance threshold. If the motion amplitude is less than or equal to the distance threshold, retain the current frame for subsequent nonlinear parameter calculation; otherwise, discard the current frame.

[0093] In practical implementation, a block motion estimation method is adopted, referring to, for example... Figure 4 As shown, the ROI region of the B-mode ultrasound image is first divided into M×N sub-blocks. Then, a concept is introduced: the image with the earlier number in the image sequence is called the previous frame, and the image with the next higher number is called the current frame (meaning the image just acquired). A block is then selected from the current frame as the target block for a motion search. Block matching criteria can include sum of absolute difference (SAD), mean square error (MSE), and normalized cross correlation function (NCCF). These methods offer similar accuracy for block matching, but their computational complexity varies significantly, with SAD requiring the least computation. The calculation method for SAD is as follows:

[0094] Formula (1)

[0095] In the above formula (1), This indicates the target image patch selected in the first frame. This represents the image patch at position (i, j) in the current frame. The smaller the SAD-based matching degree value, the greater the similarity between patches. The above formula yields M×N SAD values. The smallest SAD value is then selected and compared with the absolute difference and a threshold (e.g., 0.2). If the SAD value is less than the absolute difference and threshold, the relative motion between the two frames can be determined based on the current frame's position.

[0096] Block matching can also be achieved using the squared error of the image and cross-correlation, as shown in formulas (2) and (3) respectively:

[0097] Formula (2)

[0098] Formula (3)

[0099] In the above normalized cross-correlation formula, This indicates the target image patch selected in the first frame. This represents the image patch at position (i, j) in the current frame. Unlike the SAD standard, the higher the value of the normalized cross-correlation coefficient, the greater the correlation. Therefore, when judging the mutual motion between frames, it is necessary to find the maximum value of M×N normalized cross-correlation coefficients and compare it with the normalized cross-correlation threshold (e.g., a value of 0.7). When it is greater than the normalized cross-correlation threshold, it is considered that the image patch of the subsequent frame is formed by the motion of the target image patch of the previous frame. Finally, the position of the found image patch is used to judge the amount of motion of the image.

[0100] In the process of judging the motion amplitude, the motion relationship between the first frame image and the subsequent frame images is found based on the block matching algorithm, that is, the number of pixels that the subsequent frame deviates from the first frame. Then, the number of pixels is compared with the distance threshold. If it is greater than the preset distance threshold, the frame is discarded. If it is less than the distance threshold, the motion deviation of the current frame is considered to be within the expected range and is retained.

[0101] For example, if the similarity of the target image patch in the first frame is calculated with the subsequent M×N image patches, M×N similarity values ​​will be obtained. If the SAD (Simultaneous Average Difference) method is used, the minimum value is selected and compared with the absolute difference and a threshold. If it is less than the absolute difference and the threshold, then the target image patch is considered to have successfully matched with the subsequent frames. Then, the position index (pixel position) of the minimum SAD value is calculated. Compare the position of the target image patch with the position of the target image patch (also) To perform position calculations, you can use the following formula: The calculation is performed to determine the relative motion between the first and subsequent frames. A distance threshold is set, and if the distance calculated by the above formula is greater than the distance threshold, the relative motion is considered too large and is discarded.

[0102] 4. Perform nonlinear parameter scanning on the ROI regions in the filtered image frames to obtain a nonlinear parameter matrix. The region used for nonlinear parameter scanning can be an inscribed curvilinear trapezoid of the aforementioned ROI, such as... Figure 5 The green curved trapezoidal ROI region in the middle is enclosed by a white rectangle, which is the rectangular ROI region for motion estimation in step 3. Since the nonlinear parameter calculation is based on the IQ envelope signal of the RF radio frequency signal, the curved trapezoidal ROI region of B_mode mapped to the IQ envelope is exactly a regular rectangular region. After selecting the curved trapezoid within the ROI, the corresponding region of the tissue probability weight W from step 2 is extracted. Because the ROI is a curved trapezoid, the tissue probability weight of the rectangular region is obtained by performing an inverse scan transformation based on the parameters of the ultrasound system scan transformation. Used for subsequent Kalman filter calculations.

[0103] For frames where the normalized cross-correlation coefficient between the first and subsequent frames is greater than the normalized cross-correlation threshold and meets the distance threshold, a nonlinear scan is performed on the curved trapezoidal shape within the ROI to obtain a nonlinear parameter matrix. Simultaneously, image cross-correlation is calculated between the first frame and the next frame, following the same process. The calculated cross-correlation coefficient is compared with the normalized cross-correlation threshold and the distance threshold. If the conditions are met, a nonlinear parameter scan is initiated to obtain the nonlinear parameter matrix based on the ROI of that frame. This process continues until a predetermined time is reached, at which point a series of nonlinear parameter matrices are obtained.

[0104] like Figure 6 The diagram shows the algorithm flowchart for obtaining the nonlinear parameter matrix of the curved trapezoidal shape within each frame of the ROI under nonlinear parameter scanning. First, four sets of echo signals under high and low voltage conditions are obtained for each frame image, namely the high voltage fundamental signal, high voltage harmonic signal, low voltage fundamental signal, and low voltage harmonic signal. Then, median filtering is performed on each set of signals to eliminate noise interference. Finally, the nonlinear parameter equation is constructed, and the form of the nonlinear parameter equation is as follows:

[0105] Formula (4)

[0106] The calculation method for b in the above formula is as follows:

[0107] Formula (5)

[0108] x is defined as follows:

[0109] Formula (6)

[0110] In formula (6) , The value of is mainly affected by the probe's transmission frequency and can be estimated in advance through experiments. This represents the value related to the nonlinear parameter, specifically... , This represents the normalized voltage, specifically the ratio of the high to low input voltage.

[0111] The aforementioned nonlinear equations can be solved using numerical analysis. Initial values ​​can be generated using a Gaussian distribution, with the mean of the Gaussian distribution being the mean of the theoretical nonlinear parameters of the liver and fat. The variance should not be too large; 4 is suitable because the nonlinear parameter value of the liver parenchyma is approximately 6, and that of fat is approximately 10, with a mean of approximately 8. Therefore, choosing a variance of 4 ensures that most of the generated data is distributed within the interval (6, 10). After setting the initial values, a numerical analysis method can be chosen to solve for the parameter x. Numerical analysis methods such as Newton's method or BFGS method can be used to obtain the nonlinear parameter matrix A / B.

[0112] 5. After the above process is completed, the nonlinear parameter matrix A / B will be obtained, and its matrix form is as follows: Figure 7 As shown, Figure 7 Each pixel in the nonlinear matrix corresponds to several liver and fat cells. According to relevant studies, each pixel corresponds to a liver cell region of size 15×15. Assuming the nonlinear parameter value of lipid droplets is 9±0.5 and the nonlinear parameter value of liver parenchyma is 6.8±0.5, if the proportion of liver fat content in the specified target region is proportional to the value of the nonlinear parameter, then the numerical range of the nonlinear parameter matrix is ​​7.9±0.5.

[0113] In this invention, it is assumed that the liver fat content and the magnitude of the nonlinear parameter within the specified ROI region have a linear relationship. The linear relationship model is constructed as follows:

[0114]

[0115] f represents the fat content corresponding to each nonlinear parameter value. These are the nonlinear parameter values ​​corresponding to liver parenchyma without fat. P represents the non-linear parameter value corresponding to fat; P represents the non-linear parameter value corresponding to a certain pixel, i.e., the grayscale value.

[0116] Based on the linear relationship model, each element in the aforementioned nonlinear parameter matrix is ​​converted into a liver fat content value, which is between 0 and 1. The converted fat content matrix is ​​as follows: Figure 8As shown, in actual liver fat content detection, deviations between the measured fat content and the actual value can occur due to patient movement, cardiac traction on the liver, ultrasound probe vibration, or differences in the ROI (Region of Interest) location. Furthermore, because the liver ROI selection differs from an ideal nonlinear membrane, and the human liver contains a complex composition of various substances, the nonlinear parameter values ​​within a specified ROI will introduce some errors in the fat content measurement. Based on the above analysis, this invention proposes a Kalman filter algorithm based on multi-frame images to obtain stable and reliable liver fat content values.

[0117] 6. Kalman filtering, as a recursive Bayesian estimation method, can obtain better estimates by combining prior and observational data under dynamic processes and noise conditions. When multi-frame time-series ultrasound data is introduced, it establishes a dynamic model of the slow change in fat content over time, which can effectively improve the stability and accuracy of detection.

[0118] The first step in Kalman filtering is to establish the state equation, as shown in equation (7):

[0119] Formula (7)

[0120] In the above formula (7), Indicates noise It follows a multidimensional Gaussian distribution with a mean of 0 and a covariance matrix of Q. and Let A represent the fat fraction vector between the current and previous time steps. This vector can be obtained by flattening the liver fat content fraction matrix obtained in step 5. A is the process transition matrix. Since the fat content is consistent between frames within a short period of time, A is the identity matrix here. , This represents process noise, which has a mean of 0 and a covariance matrix of... The Gaussian distribution of the noise means that since each pixel is independent and does not affect the others, the covariance matrix of the process noise is a diagonal matrix. Since it is a diagonal matrix, in practical applications, the variance of the noise in this process should be very small. This is because the position of the ROI remains almost unchanged over a very short period of time, so the observed fat content fraction will not change in principle. Therefore, the noise variance of the state equation can be set to be small, specifically 0.01.

[0121] The observation equation is as described in formula (8):

[0122] Formula (8)

[0123] In the above formula (8), It is a nonlinear parameter vector, where H represents the measurement transfer matrix. , and These represent the nonlinear parameter values ​​of liver parenchyma and fat under ideal conditions, respectively; both are scalars. It is the identity matrix. It is the fat content score vector of the current frame. It measures noise. express The measurement noise follows a multidimensional Gaussian distribution with a mean of 0 and a covariance matrix of R. The source of this measurement noise may be nonlinear noise effects from other hardware components of the system rather than the liver itself, the influence of other substances within the liver ROI, or errors in the nonlinear parameter calculation method itself.

[0124] Kalman filtering can smooth out errors, noise, and outliers caused by various reasons, but it requires prior knowledge of the specific distribution of the errors. Here, it is assumed that the process error covariance Q is a diagonal matrix with values ​​of 0.01, and the measurement error covariance R is also a diagonal matrix with identical elements. The value of this diagonal matrix can be obtained experimentally. Given that the liver fat content of a certain size tissue region is a constant, the nonlinear parameter matrix of this tissue is measured. Ideally, this nonlinear parameter matrix should be a constant matrix, but noise exists during the measurement process, so the value of this nonlinear parameter matrix will definitely be different. Next, the variance of this nonlinear parameter matrix is ​​calculated and recorded. The nonlinear parameter matrix is ​​calculated and the variance is recorded again in regions with different fat contents. Finally, the mean of all recorded variances is taken as the variance in the process of nonlinear parameter measurement of fat content.

[0125] After establishing the state equations for the Kalman filter, initialize the error matrix, which is the actual error between the true and predicted values. This value can be randomly set; for example, a diagonal matrix with the same dimensions as the nonlinear parameter matrix, with a size of 0.1. The fat content score vector can be directly derived from the nonlinear parameter matrix of the first frame; for example, the fat content score vector can be obtained by directly flattening the fat content score matrix from step 5. The Kalman filter consists of two steps: prediction and update. The prediction is as follows:

[0126] Formula (9)

[0127] This is the posterior estimate of the previous frame. If it's based on the prediction of the first frame, then this value is the initial fat content score calculated from the nonlinear parameter matrix of the first frame. Here, A is the identity matrix I, indicating that the fat content is consistent between the previous and subsequent frames. This refers to the prior fat content derived from the equation of state. The prior error covariance, representing the uncertainty of the prediction, is:

[0128] Formula (10)

[0129] In the above formula, A is the state transition matrix, which is the identity matrix I in this case. and Let represent the prior error covariance matrix and the posterior error covariance matrix at time k-1. If it is the first frame, then... Q is the initialization error matrix mentioned above, and Q is the covariance matrix of the process model.

[0130] The update steps are as follows:

[0131] Formula (11)

[0132] It is the optimal estimate of the fat content fraction matrix. This represents the observed nonlinear parameter matrix at the current moment. It is the prior estimate of liver fat content (that is, the initial value or the optimal estimate from the previous step), which is the value calculated by formula (9), where H is the measurement state matrix. K is the Kalman gain, and its calculation equation is as described in formula (12):

[0133] Formula (12)

[0134] In the above formula, H is the prior error covariance matrix calculated by formula (10), and H is the measurement transition matrix. Let R be the transpose matrix, and R be the covariance matrix of the measurement equation. However, this invention will modify this matrix, and the modified equation is as follows:

[0135] Formula (13)

[0136] In formula (13) This is the organization probability weight probability map corresponding to the inscribed curvilinear trapezoid of the ROI calculated using the nonlinear parameters in step 4 above. The parameter is very small. Its main purpose is to ensure that the denominator of the above formula is not zero. After the weighting of formula (13), when the probability of a region being liver parenchyma is small, the covariance of the corresponding region in R will increase, thereby reducing the influence weight of the predicted value of the nonlinear parameter on the final result.

[0137] The update formula for the posterior error covariance matrix is:

[0138] Formula (14)

[0139] In the above formula, I is the identity matrix, and K is the Kalman gain. It is the prior error covariance matrix.

[0140] After the update step is completed, the posterior error covariance matrix and the optimal estimate from the update step are used as the prediction values ​​for the new round. After all frames have been iterated, the optimal estimate of the final output is... The average value is the final liver fat content score. This method can effectively reduce the impact of noise on the measurement results. Compared with directly calculating the average liver fat content of multiple frames, this method can effectively suppress the influence of non-liver parenchyma areas on the final result and outliers caused by various factors.

[0141] Based on the same inventive concept, this invention also provides a multi-frame image fatty liver content detection system based on motion inhibition. Since the principle underlying this system is similar to the aforementioned multi-frame image fatty liver content detection method based on motion inhibition, the implementation of this system can refer to the implementation of the aforementioned method; repeated details will not be elaborated further. (Refer to...) Figure 9 As shown, the system includes:

[0142] The ultrasound imaging module is used to select the region of interest in the patient's liver area and set the processing time; it acquires liver images in real time under B-mode ultrasound scanning; it realizes standardized localization and data acquisition of the target area, providing high-quality basic image data for subsequent processing, and ensuring the standardization and repeatability of the detection.

[0143] The liver parenchyma probability weight calculation module is used to generate a liver parenchyma probability weight distribution map for each frame of image. This can effectively reduce the interference of non-target tissues on the detection results, provide key weight basis for subsequent Kalman filtering, and improve the accuracy of measurement.

[0144] The motion estimation module is used to perform inter-frame motion estimation and filter out image frames whose motion amplitude is within a preset threshold. It fundamentally suppresses motion artifacts caused by breathing, probe shaking, or heartbeat, ensuring spatial consistency of multi-frame data used for analysis and laying a solid foundation for accurate fusion.

[0145] The nonlinear parameter calculation module is used to perform nonlinear parameter scanning on the filtered image frames to obtain the nonlinear parameter matrix.

[0146] The fat content conversion module is used to convert each nonlinear parameter value in the nonlinear parameter matrix into a liver fat content score value, thereby obtaining a liver fat content score matrix. This module realizes the mapping from physical parameters to clinically understandable fat percentage content, generating an intuitive fat distribution score matrix.

[0147] The Kalman filtering module applies the Kalman filtering algorithm to perform fusion filtering on the liver fat content fraction matrix to obtain the final quantitative value of liver fat content. The multi-frame fat content fraction matrix, liver parenchyma probability weights, and motion estimation information generated by the aforementioned module are fused, and the optimal estimate is obtained through prediction and update iterations.

[0148] This system, through its integrated modular design, transforms innovative multi-frame fusion and anti-motion interference algorithms into a practically operable medical device system. Ultimately, it achieves non-invasive, accurate, and stable quantitative detection of liver fat content, solving the core pain points of traditional clinical methods, such as high trauma, high cost, inability to quantify, or unstable results.

[0149] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0150] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those 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 invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting fatty liver content in multi-frame images based on motion inhibition, characterized in that, Includes the following steps: Select the region of interest in the patient's liver area and set the processing time; acquire liver images in real time under B-mode ultrasound scanning; Generate a probability weight distribution map of liver parenchyma for each frame of image; Perform inter-frame motion estimation and filter out image frames whose motion amplitude is within a preset threshold; The selected image frames are scanned using nonlinear parameters to obtain the corresponding nonlinear parameter matrix; Each nonlinear parameter value in each of the aforementioned nonlinear parameter matrices is converted into a liver fat content score value to obtain the corresponding liver fat content score matrix. The Kalman filter algorithm is applied to fuse and filter multiple liver fat content fraction matrices to obtain the final quantitative value of liver fat content. Specifically, this includes: establishing state equations and observation equations; Initialize the state vector and error covariance matrix; perform prediction and update steps, adjust the measurement noise covariance by combining the probability weights of liver parenchyma; output the mean of the optimal fat content score matrix as the final liver fat content.

2. The method for detecting fatty liver content in multi-frame images based on motion inhibition according to claim 1, characterized in that, The probability weight distribution map of the liver parenchyma is generated by a deep learning semantic segmentation model. The model adopts a U-Net network structure, and during training, only pixels in the liver parenchyma region are labeled. Non-liver parenchyma regions include blood vessels, shadows, calcifications and other non-liver tissues. The output is the probability value of each pixel belonging to the liver parenchyma.

3. The method for detecting fatty liver content in multi-frame images based on motion inhibition according to claim 1, characterized in that, Perform inter-frame motion estimation and select image frames whose motion amplitude is within a preset threshold; including: The first frame captured in the image sequence is set as the reference frame, and the subsequent frames are set as the current frames in sequence. The regions of interest in the reference frame and the current frame are respectively divided into multiple image sub-blocks of size M×N pixels; In the current frame, for the target image sub-block in the reference frame, block matching calculation is performed within the corresponding search area to find the optimal matching sub-block; The relative motion amplitude between the current frame and the reference frame is calculated based on the positional difference between the target image sub-block and the optimal matching sub-block. The motion amplitude is compared with a distance threshold. If the motion amplitude is less than or equal to the distance threshold, the current frame is retained for subsequent nonlinear parameter calculation; otherwise, the current frame is discarded.

4. The method for detecting fatty liver content in multi-frame images based on motion inhibition according to claim 3, characterized in that, The matching criteria used in the block matching calculation are the sum of absolute differences, squared errors, or normalized cross-correlation functions.

5. The method for detecting fatty liver content in multi-frame images based on motion inhibition according to claim 1, characterized in that, The selected image frames are subjected to a nonlinear parameter scan to obtain the corresponding nonlinear parameter matrix; including: Four sets of echo signals were acquired: high-voltage fundamental, high-voltage harmonic, low-voltage fundamental, and low-voltage harmonic. Perform median filtering on the signal; Construct and solve the nonlinear parametric equations to obtain the corresponding nonlinear parameter matrix.

6. The method for detecting fatty liver content in multi-frame images based on motion inhibition according to claim 1, characterized in that, Each nonlinear parameter value in each of the aforementioned nonlinear parameter matrices is converted into a liver fat content score value to obtain the corresponding liver fat content score matrix; including: The nonlinear parameter values ​​for obtaining the ideal liver parenchyma are: Nonlinear parameter values ​​of fat : Based on the linear relationship between the nonlinear parameter values ​​and fat content in a specified region of the liver, the following linear relationship model is constructed: f The fat content corresponding to each nonlinear parameter value These are the nonlinear parameter values ​​corresponding to liver parenchyma without fat. P represents the non-linear parameter value corresponding to fat; P represents the non-linear parameter value corresponding to a certain pixel, i.e., the grayscale value. Based on the linear relationship model, each element in each nonlinear parameter matrix is ​​converted into a liver fat content score between 0 and 1, resulting in a liver fat content score matrix.

7. A multi-frame image fatty liver content detection system based on motion inhibition, characterized in that, The system, employing the method of any one of claims 1-6, comprises: The ultrasound imaging module is used to select the region of interest in the patient's liver area and set the processing time; it acquires liver images in real time under B-mode ultrasound scanning. The liver parenchyma probability weight calculation module is used to generate a liver parenchyma probability weight distribution map for each frame of image; The motion estimation module is used to perform inter-frame motion estimation and filter out image frames whose motion amplitude is within a preset threshold. The nonlinear parameter calculation module is used to perform nonlinear parameter scanning on the filtered image frames to obtain the corresponding nonlinear parameter matrix. The fat content conversion module is used to convert each nonlinear parameter value in each of the nonlinear parameter matrices into a liver fat content score value, thereby obtaining the corresponding liver fat content score matrix. The Kalman filter module applies the Kalman filter algorithm to perform fusion filtering on multiple liver fat content fraction matrices to obtain the final quantitative value of liver fat content.

Images