A method and system for automatically evaluating liver stiffness based on abdominal CT images

Abstract

The present invention relates to the field of medical image processing, and discloses a method and system for automatically evaluating the degree of liver stiffness based on abdominal CT images, including: preprocessing the CT image, segmenting the preprocessed CT image into multiple patches; constructing a neural network model, extracting a training data set, training the network model, inputting the segmented CT image, and outputting the reasoning result of each patch; merging the reasoning result of each patch into the reasoning result of the liver volume in the whole image; obtaining the edge of the liver in the CT image based on the liver volume reasoning result; performing LSN measurement based on the edge of the liver; and evaluating the degree of liver stiffness based on the LSN measurement result. The present invention automatically identifies the liver and reasonable liver edges based on AI, and the whole process is automatically processed by the system, eliminating the need for professional physicians, pre-use training of the software, and cumbersome operating procedures.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a method and system for automatically evaluating liver stiffness based on abdominal CT images. Background Art

[0002] Cirrhosis is a type of liver structural and functional abnormality caused by the long-term development of various chronic liver diseases (such as viral hepatitis, alcoholic liver disease, autoimmune liver disease, etc.), which manifests as extensive necrosis of liver cells, diffuse proliferation of liver fibrosis, and formation of pseudolobules, which eventually lead to gradual deformation and hardening of the liver. During the course of cirrhosis, liver inflammation gradually worsens, eventually leading to liver decompensation, and a series of serious complications such as jaundice, ascites, and portal hypertension.

[0003] Liver cirrhosis is not only an important manifestation of liver disease itself, but also an important risk factor for hepatocellular carcinoma. Therefore, accurate assessment of the degree of liver cirrhosis is crucial for the diagnosis of the disease, formulation of treatment plans and prognosis. There are currently several main methods for assessing the degree of liver cirrhosis:

[0004] Liver biopsy: Liver biopsy is the gold standard for diagnosing the degree of liver cirrhosis. It extracts a small amount of liver tissue through percutaneous liver puncture, and observes the changes in its tissue morphology under a microscope to make a corresponding diagnosis. Although this method is accurate, it is invasive and will cause physical and psychological pain to patients. It is not suitable for all patients, and the cost is high and the results are relatively slow.

[0005] Serological marker detection: Serological marker detection includes detection of liver function indicators such as total bilirubin, direct bilirubin, aspartate aminotransferase, alanine aminotransferase, albumin, globulin, glutamyl transpeptidase, alkaline phosphatase, and four liver fibrosis items. These indicators can reflect the functional status and degree of fibrosis of the liver, thereby indirectly assessing the degree of cirrhosis. However, this method is easily affected by many factors, such as drugs, diet, infection, etc., and is therefore not accurate enough.

[0006] Imaging evaluation: Imaging evaluation includes ultrasound, CT, MRI, etc. These methods can intuitively show the size, shape, echo distribution and presence of nodules of the liver, so as to roughly judge the degree of cirrhosis. However, imaging evaluation is extremely dependent on the doctor's experience and subjective judgment, sometimes there is a large subjective deviation, and it is difficult to standardize and quantify.

[0007] CT, as a non-invasive and intuitive imaging assessment method, can more intuitively show the incomplete and jagged changes in the liver edge caused by liver surface nodules in cirrhosis. These surface nodules cause the liver to change from a smooth surface in normal morphology to a surface with multiple ridges and bumps. The difference between a smooth liver surface and a liver surface with nodules is as follows: Figure 1 shown.

[0008] Therefore, the possibility of cirrhosis can be assessed based on the smoothness of the liver surface in CT. In 2015, Andrew D. Smith proposed a method for assessing the degree of cirrhosis based on the non-smoothness of the liver edge manually outlined by function fitting, the Liver Surface Nodularity Quantification (LSN). This method involves trained doctors manually outlining a segment of the liver edge that may show characteristics of cirrhosis on dedicated software, and fitting it with a high-order function to calculate the degree of liver non-smoothness of the edge segment to assess the degree of cirrhosis of the liver. This method transforms the image-based assessment of cirrhosis from being completely dependent on the doctor's subjective judgment to a partially quantifiable operation. In 2018, Tae-Hoon Kim further verified the effectiveness of LSN in simple steatosis and non-alcoholic fatty liver disease. In general, LSN improves the objectivity of non-invasive testing, but there are still problems in its application, such as complex and time-consuming operation and the need for trained experts to operate. Summary of the invention

[0009] The main purpose of the present invention is to solve the technical problems in the prior art that the operation is complicated and time-consuming and requires trained experts to operate. A method for automatically evaluating liver stiffness based on abdominal CT images comprises the following steps:

[0010] Acquire abdominal CT images, and extract liver edge stereo images from the abdominal CT images;

[0011] Obtaining the coordinate point with the smallest x value and y value in the coordinates of the single-layer liver edge image in the liver edge stereo image;

[0012] Connect the coordinate points with the smallest x and y values ​​in the single-layer liver edge image coordinates to obtain a continuous edge mask;

[0013] Calculate smooth reference lines for edge masks;

[0014] The LSN of a single-layer liver edge image is calculated using the following formula:

[0015]

[0016] in:

[0017] di represents the vertical distance from the i-th edge boundary point to the smooth reference line;

[0018] N is the total number of sampling points on the liver surface;

[0019] The degree of liver stiffness was evaluated based on the LSN calculation results of all layers in the liver edge stereo images.

[0020] The step of extracting a liver edge stereoscopic image from an abdominal CT image comprises:

[0021] Preprocess the CT image and divide the preprocessed CT image into multiple patches;

[0022] Build a neural network model, extract training data sets, train the network model, input segmented CT images, and output the inference result of whether each patch is liver tissue;

[0023] According to the inference result, a liver edge stereo image is acquired.

[0024] The preprocessing of the CT image and segmenting the preprocessed CT image into a plurality of patches includes:

[0025] Input the original abdominal plain CT image;

[0026] The CT image brightness HU value was limited to a maximum value of 260 and a minimum value of -25, and then normalized;

[0027] Scale the CT image so that each voxel in its voxel space meets a fixed size in x, y, and z;

[0028] The CT image is segmented into multiple patches using a sliding window with a step size of a fixed patch side length.

[0029] The step of obtaining a liver edge stereoscopic image according to the inference result includes:

[0030] According to the inference results of each patch, all the inference results are weighted by three-dimensional Gaussian, so that when the patches are merged, the weight of the inference result of each patch is the highest at the center and the lowest at the edge, and then merged into the inference result of the liver volume in the whole image.

[0031] An automatic liver stiffness assessment system based on abdominal CT images, the system comprising:

[0032] An extraction unit, which acquires an abdominal CT image and extracts a liver edge stereoscopic image from the abdominal CT image;

[0033] The edge mask screening unit obtains the coordinate point with the smallest x value and y value of the single-layer liver edge image in the liver edge stereo image; connects the coordinate point with the smallest x value and the coordinate point with the smallest y value in the single-layer liver edge image to obtain a continuous edge mask;

[0034] The single-layer LSN calculation unit calculates the smooth reference line of the edge mask and uses the following formula to calculate the LSN of the single-layer liver edge image:

[0035]

[0036] in:

[0037] di represents the vertical distance from the i-th edge boundary point to the smooth reference line;

[0038] N is the total number of sampling points on the liver surface;

[0039] The evaluation unit evaluates the liver stiffness degree according to the LSN calculation results of all layers in the liver edge stereo image.

[0040] A third aspect of the present invention provides an electronic device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned method for automatically assessing the degree of liver stiffness based on abdominal CT images.

[0041] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned method for automatically assessing liver stiffness based on abdominal CT images.

[0042] The present invention has the following beneficial effects:

[0043] The present invention automatically identifies the liver and reasonable liver margins based on AI, and the entire process is handled automatically by the system, eliminating the need for professional doctors, software training and cumbersome operating procedures.

[0044] The present invention recognizes the liver and the edge based on AI, and is not limited by many conditions when outlining the edge in the semi-automatic method. It can also achieve stable edge recognition results when dealing with fat, ascites, and close to the edge of the trunk.

[0045] The single measurement of the present invention generally takes about 0.5 to 1.5 minutes, and can be carried out in batches at the same time without human operation, which greatly improves the efficiency of use. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is the difference between a smooth liver surface and a liver surface with nodules;

[0047] Figure 2 The present invention provides a flowchart of a method for automatically evaluating liver stiffness based on abdominal CT images. DETAILED DESCRIPTION

[0048] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0049] Normal liver usually has uniform density and clear and smooth liver edges. However, liver cirrhosis will show uneven liver edges on imaging, especially in the decompensated stage. This is caused by multiple isodense regenerative nodules in the liver. Therefore, it is theoretically feasible to evaluate liver cirrhosis based on liver surface nodules, that is, the smoothness of the liver surface.

[0050] At present, no patents for automatic quantification of liver surface nodules (LSN) based on CT for evaluating liver cirrhosis have been found, but similar implementation schemes have been published in foreign journals. The specific steps of this method are as follows:

[0051] 1. The operator first zooms the CT image to the liver area on the dedicated software.

[0052] 2. Use the brush to roughly outline a section of the liver edge area that is representative of the liver surface nodules. During this process, the liver edge outlined by the brush should be careful to avoid:

[0053] a) Fat edges

[0054] b) There is a margin of ascites

[0055] c) Close to the edge of the torso

[0056] d) There are sharp bends or fissures at the edge of the liver

[0057] 3. Expand the area outlined by the brush up and down to obtain slices of the liver edge area near the 5-layer CT.

[0058] 4. The software performs semi-automatic liver edge recognition based on the edge areas of these five layers

[0059] 5. Perform polynomial fitting on the five extracted liver edges and calculate the differences between these edges and the polynomial fitting results.

[0060] 6. Assess the patient’s degree of cirrhosis based on the size of the difference.

[0061] First of all, doctors need to undergo professional training and have a certain number of years of experience. And although there is function fitting, the specific selection of which edge to outline requires doctors to have a full understanding of cirrhosis and the edge extraction method of the software. For example, doctors need to find edges with obvious characteristics of cirrhosis while avoiding outlining areas with fat, ascites, close to the torso, and natural sharp turns or fissures. Secondly, the process of manually outlining the edge of the liver and fitting a function is relatively cumbersome and not suitable for batch operations on a large number of examinations. Finally, the edges outlined by different doctors or the same doctor at different times may be different, resulting in unstable results and sometimes difficult to reproduce the evaluation results.

[0062] 1. With respect to the above-mentioned prior art, analysis shows that the prior art has the following disadvantages:

[0063] a) The operator must be a professional physician with a certain number of years of experience, so that he or she can locate the liver area on the CT image and outline the liver edge that is more representative of the liver surface nodules.

[0064] b) Operators must be specially trained to operate the software and avoid a large number of situations that the software cannot handle as mentioned above.

[0065] c) Due to the above-mentioned limitations, manually outlining the edge area is rather cumbersome. According to statistics in the paper, professional doctors can only perform one measurement at a time using this method, and the time required for a single LSN measurement is generally between 0.5 and 6 minutes. Therefore, the labor cost is high and it is not suitable for batch operations on a large number of examinations.

[0066] d) Finally, since the edges for calculation are manually outlined by doctors, the edges outlined by different doctors or the same doctor at different times may be very different, resulting in unstable LSN calculation results.

[0067] 2. Compared with the prior art, the present invention solves the above problems in a targeted manner based on technologies such as artificial intelligence and automation.

[0068] a) Automatic identification of the liver and reasonable liver margins based on AI, with the entire process handled automatically by the system, eliminating the need for professional physicians, pre-software training and cumbersome operating procedures.

[0069] b) AI-based liver and edge recognition is free from the many restrictions of semi-automatic edge delineation. Stable edge recognition results can be obtained even when dealing with fat, ascites, and close to the trunk edge.

[0070] c) A single measurement generally takes about 0.5 to 1.5 minutes, and can be performed in batches at the same time without human operation. The efficiency of use is greatly improved.

[0071] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 2 The first embodiment of the method for automatically evaluating liver stiffness based on abdominal CT images in the embodiment of the present invention includes:

[0072] 1. Preprocess the CT image and divide the preprocessed CT image into multiple patches;

[0073] Input the original plain scan CT image of the abdomen; limit the HU value of the CT image brightness to a maximum value of 260 and a minimum value of -25, and then normalize it according to a fixed HU value mean and a fixed standard deviation; scale the CT image so that each voxel in its voxel space meets a fixed size in x, y, and z; use a sliding window with a step size of a fixed patch side length to segment the CT image into multiple patches.

[0074] 2. Build a neural network model, extract training data sets, train the network model, input segmented CT images, and output the inference results of each patch;

[0075] According to the inference results of each patch, all the results are Gaussian weighted and merged into the inference result of the liver volume in the whole image.

[0076] 3. Based on the liver volume inference results, obtain the edge of the liver in the CT image;

[0077] The mask corresponding to the liver volume is expanded by single pixel, and the original mask is removed from the expanded result to obtain the edge mask.

[0078] 4. Measure LSN based on the edge of the liver;

[0079] Obtain the coordinate point with the smallest x value and the coordinate point with the smallest y value in the mask; connect the coordinate point with the smallest x value and the coordinate point with the smallest y value in the mask to obtain a continuous edge mask; use LSN to measure the edge mask obtained in the previous step.

[0080] The LSN calculation method is as follows:

[0081] Determine liver surface reference line

[0082] Select the long axis of the liver (usually along the anatomical direction of the liver).

[0083] Create a smooth reference line on the liver surface (you can use B-spline fitting or other smoothing methods).

[0084] This reference line can be regarded as an ideal “smooth liver surface.” The edge mask used in the present invention is obtained by connecting the coordinate point with the smallest x value and the coordinate point with the smallest y value in the mask.

[0085] Calculating surface irregularities

[0086] Point-to-curve distance measurement: Calculate the vertical distance from each point on the liver surface to the reference line.

[0087] Curvature change rate: The local change in the curvature of the liver surface is calculated. Areas with larger curvature changes usually indicate more obvious surface nodules.

[0088] Statistical analysis:

[0089] Calculate statistical indicators such as mean, standard deviation, median, etc. for all points.

[0090] The proportion of points whose distance exceeds a certain threshold (such as 2mm) is counted.

[0091] Calculate the LSN index

[0092] The LSN calculation formula is:

[0093]

[0094] in:

[0095] di represents the vertical distance from the i-th boundary point to the smooth reference line;

[0096] N is the total number of sampling points on the liver surface.

[0097] 5. Evaluate the degree of liver stiffness based on the LSN measurement results.

[0098] Assessment of liver stiffness

[0099] The relationship between LSN and liver fibrosis grade

[0100] According to the study, the LSN index is correlated with the METAVIR liver fibrosis score (F0-F4):

[0101] F0-F1 (normal or mild fibrosis): LSN values ​​are small (usually <2.5).

[0102] F2 (moderate fibrosis): LSN values ​​are between 2.5-3.5.

[0103] F3 (severe fibrosis): LSN value is 3.5-4.5.

[0104] F4 (cirrhosis): LSN value >4.5, the liver surface is highly irregular and presents obvious nodules.

[0105] Specifically, the scheme of the present invention is as follows:

[0106] Preprocessing of CT images: After inputting the original plain scan CT images of the abdomen, the original images were normalized and the voxel size was adjusted. The image brightness (HU value) was limited to a range of 260 for the maximum value and -25 for the minimum value, and then normalized according to the mean of 100 and the standard deviation of 50. After processing, the images were scaled so that each voxel in the voxel space (voxel spacing) met the size of (0.75mm, 0.75mm, 1.0mm) in x, y, and z. A sliding window with a step size of 0.5 was used to segment the CT images into multiple patch sizes of (192, 192, 64) for subsequent liver volume segmentation.

[0107] Liver volume segmentation: Use a 3D U-net neural network to input the preprocessed image and output the inference results of each patch. All results are Gaussian weighted and merged into the inference result of the liver volume in the entire image.

[0108] The calculation method of the three-dimensional Gaussian weighting is as follows:

[0109] Step 1: Define kernel size

[0110] The kernel size is usually an odd number (such as 5x5x5) and is symmetrical about the center point.

[0111] The nuclear radius k is determined by σ, and is generally taken as k = 3σ (covering 99.7% of the energy).

[0112] Step 2: Calculate the weight of each voxel

[0113] For each coordinate (i, j, k) in the kernel, calculate the weight:

[0114]

[0115] Here, c is the coordinate of the center of the kernel (e.g., c=2 in a 5x5x5 kernel).

[0116] Normalizing weights normalizes all weights so that the sum is 1:

[0117] .

[0118] The 3D U-net neural network model needs to be trained before use. The neural network used for inference in this patent is trained using public data after being fine-tuned by experts. During the training process, linear scaling, random rotation, gamma transformation and other operations are used for data enhancement. Dice Loss + Cross Entropy Loss is used as the loss function.

[0119] Liver edge extraction: Based on the liver volume inference result, the edges of the liver in each layer are obtained. The specific operation is to expand the mask corresponding to the liver volume by a single pixel, and remove the original mask from the expanded result to obtain the liver edge mask.

[0120] Screening effective edges: By extracting the right front edge corresponding to the left inner lobe of the liver to the right posterior lobe of the liver in each layer, those liver edges that should not be used to calculate LSN are excluded, such as the edge of the liver and blood vessels, the edge of the liver fissure, the edge of the liver turning point, the edge of the posterior part of the left and right lobes of the liver, and the edge of the caudate lobe of the liver. Because the right front edge from the left inner lobe of the liver to the right posterior lobe of the liver is continuous and smooth in a normal liver, without the existence of fissures.

[0121] The specific extraction method is to extract a continuous edge mask from the most front mask point to the most right mask point in each layer of liver edge mask. That is, a continuous edge mask connected by the coordinate point with the smallest x value and the coordinate point with the smallest y value in each layer of mask. This edge section strictly corresponds to the right front edge of the left inner lobe to the right posterior lobe of the liver in the image. The LSN of this edge section has a more significant reflection of liver sclerosis.

[0122] The above describes the method for automatically evaluating the degree of liver stiffness based on abdominal CT images in an embodiment of the present invention. The following describes the device for automatically evaluating the degree of liver stiffness based on abdominal CT images in an embodiment of the present invention:

[0123] A preprocessing unit preprocesses the CT image and divides the preprocessed CT image into multiple patches;

[0124] The inference unit builds a neural network model, extracts training data sets, trains the network model, inputs segmented CT images, and outputs the inference results of each patch;

[0125] The synthesis unit combines the inference results of each patch into the inference result of the liver volume in the whole image;

[0126] An edge extraction unit, which obtains the edge of the liver in the CT image based on the liver volume inference result;

[0127] The LSN measurement unit performs LSN measurement according to the edge of the liver;

[0128] The assessment unit assesses the degree of liver stiffness based on the LSN measurement result.

[0129] An embodiment of the present invention also provides an electronic device, which may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) (for example, one or more processors) and memories, and one or more storage media for storing applications or data (for example, one or more mass storage devices). Among them, the memory and the storage medium may be short-term storage or permanent storage. The program stored in the storage medium may include one or more modules, each module may include a series of instruction operations in the electronic device. Furthermore, the processor may be configured to communicate with the storage medium and execute a series of instruction operations in the storage medium on the electronic device.

[0130] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input and output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that the electronic device structure in this embodiment does not constitute a limitation on the electronic device, and may include more or fewer components, or combine certain components, or arrange components differently.

[0131] An embodiment of the present invention provides a structure of an electronic device, which may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) (for example, one or more processors) and memories, and one or more storage media for storing applications or data (for example, one or more mass storage devices). Among them, the memory and the storage medium can be short-term storage or permanent storage. The program stored in the storage medium may include one or more modules, each module may include a series of instruction operations in the electronic device. Furthermore, the processor can be configured to communicate with the storage medium and execute a series of instruction operations in the storage medium on the electronic device.

[0132] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input and output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that the electronic device structure does not constitute a limitation on the electronic device, and may include more or less components than the aforementioned, or combine certain components, or arrange the components differently.

[0133] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium, and when the instructions are executed on a computer, the computer executes the steps of the aforementioned method.

[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0135] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0136] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically evaluating liver stiffness based on abdominal CT images, characterized in that: The following steps are involved: Acquire abdominal CT images, and extract liver edge stereo images from the abdominal CT images; Obtaining the coordinate point with the smallest x value and y value in the coordinates of the single-layer liver edge image in the liver edge stereo image; Connect the coordinate points with the smallest x and y values ​​in the single-layer liver edge image coordinates to obtain a continuous edge mask; Calculating a smooth reference line of the edge mask; The LSN of a single-layer liver edge image is calculated using the following formula: in: d i Represents the vertical distance from the i-th edge boundary point to the smooth reference line; N is the total number of sampling points on the liver surface; The degree of liver stiffness was evaluated based on the LSN calculation results of all layers in the liver edge stereo images.

2. The method for automatically evaluating liver stiffness based on abdominal CT images according to claim 1, characterized in that: The step of extracting a liver edge stereoscopic image from an abdominal CT image comprises: Preprocess the CT image and divide the preprocessed CT image into multiple patches; Build a neural network model, extract training data sets, train the network model, input segmented CT images, and output the inference result of whether each patch is liver tissue; According to the inference result, a liver edge stereo image is acquired.

3. The method for automatically evaluating liver stiffness based on abdominal CT images according to claim 2, characterized in that: The preprocessing of the CT image and segmenting the preprocessed CT image into a plurality of patches includes: Input the original abdominal plain CT image; Normalize the brightness of CT images; Scale the CT image so that each voxel in its voxel space meets a fixed size in x, y, and z; The CT image is segmented into multiple patches using a sliding window with a step size of a fixed patch side length.

4. The method for automatically evaluating liver stiffness based on abdominal CT images according to claim 3, characterized in that: The step of obtaining a liver edge stereoscopic image according to the inference result includes: According to the inference results of each patch, all the inference results are weighted by three-dimensional Gaussian, so that when merging patches, the weight of the inference result of each patch is the highest at the center and the lowest at the edge; Merge into the inference result of liver volume in the whole image.

5. An automatic liver stiffness assessment system based on abdominal CT images, characterized in that: The system comprises: An extraction unit, which acquires an abdominal CT image and extracts a liver edge stereoscopic image from the abdominal CT image; The edge mask screening unit obtains the coordinate point with the smallest x value and y value in the coordinates of the single-layer liver edge image in the liver edge stereo image; connects the coordinate points with the smallest x value and y value in the coordinates of the single-layer liver edge image to obtain a continuous edge mask; The single-layer LSN calculation unit calculates the smooth reference line of the edge mask and uses the following formula to calculate the LSN of the single-layer liver edge image: in: di represents the vertical distance from the i-th edge boundary point to the smooth reference line; N is the total number of sampling points on the liver surface; The evaluation unit evaluates the liver stiffness degree according to the LSN calculation results of all layers in the liver edge stereo image.

6. An electronic device, comprising a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes each step of the method for automatically assessing liver stiffness based on abdominal CT images as described in any one of claims 1 to 4.

7. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the method for automatically assessing liver stiffness based on abdominal CT images as described in any one of claims 1 to 4 are implemented.

Images