Liver fat detection method and device based on image processing, equipment and medium

Abstract

The invention relates to the technical field of image processing, and discloses a liver fat detection method and device based on image processing, equipment and a medium. Firstly, a to-be-analyzed liver plain-scan image is obtained; then, liver segmentation processing is carried out based on the to-be-analyzed liver plain scanning image, and an initial liver segmentation result is obtained; then, key region extraction is carried out based on the initial liver segmentation result, and a target key measurement region is determined; and finally, performing liver fat detection based on the target key measurement area to obtain a fat detection result corresponding to the target key measurement area. Compared with liver enhancement examination, the speed of obtaining the liver plain-scan image through the liver plain-scan examination is higher, and the radiation dose suffered by the patient is also greatly reduced. According to the method, the liver segmentation processing is performed based on the to-be-analyzed liver plain scan image to obtain the initial liver segmentation result, so that the influence of blood vessels on the analysis and measurement accuracy is effectively reduced, the accuracy of the segmentation result is improved, and the accuracy of liver fat detection is improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a liver fat detection method, device, equipment and medium based on image processing. Background Art

[0002] The liver, as the largest gland and solid organ in the human body, is not only the place where metabolism is most active, but also the site where various diseases are prone to occur.

[0003] In the related art, liver enhancement images obtained by enhanced CT examination can be used to segment the liver and further detect the fat content in each segment. However, when performing enhanced CT examination, the patient needs to bear a certain radiation dose, and the examination takes a relatively long time. Summary of the invention

[0004] The embodiments of this specification are intended to solve at least one of the technical problems in the related art to a certain extent. To this end, the embodiments of this specification propose a liver fat detection method, device, equipment and medium based on image processing.

[0005] This specification provides a method for detecting liver fat based on image processing, the method comprising:

[0006] Acquire a plain scan image of the liver to be analyzed;

[0007] Performing liver segmentation processing based on the liver plain scan image to be analyzed to obtain an initial liver segmentation result;

[0008] Extract key areas based on the initial liver segmentation results to determine target key measurement areas;

[0009] Liver fat detection is performed based on the target key measurement area to obtain a fat detection result corresponding to the target key measurement area.

[0010] In one embodiment, the target key measurement area is determined by:

[0011] Extracting key regions from the initial liver segmentation result to obtain the target key measurement region; or

[0012] A selection operation of a user based on the initial liver segmentation result or an initial key measurement region is received, and the target key measurement region is determined based on the selection operation, wherein the initial key measurement region is obtained by extracting a key region from the initial liver segmentation result.

[0013] In one embodiment, extracting key regions from the initial liver segmentation result to obtain the initial key measurement region includes:

[0014] Perform contour extraction based on the initial liver segmentation result to obtain a segmented contour;

[0015] The segmented contour is shrunk inward by a preset distance to eliminate interference from the blood vessel region, thereby obtaining the initial key measurement region.

[0016] In one embodiment, extracting key regions from the initial liver segmentation result to obtain the initial key measurement region includes:

[0017] Determining partition key points according to the initial liver segmentation result;

[0018] The initial liver segmentation result is calculated based on the partition key point as the center according to the preset volume ratio to obtain the initial key measurement area.

[0019] In one embodiment, the method further comprises:

[0020] Extracting key regions from the initial liver segmentation result to obtain initial key measurement regions;

[0021] The initial key measurement area is divided into regions to obtain a plurality of inscribed geometric bodies and use them as the target key measurement area.

[0022] In one embodiment, the initial liver segmentation result is obtained by a plain scan liver segmentation model, and the plain scan liver segmentation model is obtained by:

[0023] Acquire a sample liver plain scan image and a sample liver enhanced image; wherein the sample liver plain scan image corresponds to a plain scan image liver segmentation label, and the sample liver enhanced image corresponds to an enhanced image liver segmentation label, an enhanced image blood vessel segmentation label, and a sample key point heat map;

[0024] Inputting the sample liver plain scan image and the sample liver enhanced image into the plain scan liver segmentation model to be trained, and outputting the plain scan image liver segmentation target result, the plain scan image liver segmentation rough segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map;

[0025] Based on the plain scan image liver segmentation label, the enhanced image liver segmentation label, the enhanced image blood vessel segmentation label, the sample key point heat map, the plain scan image liver segmentation target result, the plain scan image liver segmentation coarse segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map, the parameters of the plain scan liver segmentation model to be trained are updated to obtain a trained plain scan liver segmentation model.

[0026] In one embodiment, the sample key point heat map is obtained by:

[0027] Performing liver segmentation and blood vessel segmentation processing on the sample liver enhanced image to obtain a liver mask image, wherein the liver mask image includes a liver segmentation mask and a blood vessel segmentation result mask;

[0028] Extracting vascular bifurcation positions and key anatomical positions based on the liver mask image to obtain key point position information;

[0029] The sample key point heat map is generated based on the key point position information.

[0030] In one embodiment, the sample liver plain scan image and the sample liver enhanced image are input into the training

[0031] The plain scan liver segmentation model outputs the plain scan image liver segmentation target result, the plain scan image liver segmentation rough segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map, including:

[0032] Using the plain scan liver segmentation model to be trained, coarse-grained detection is performed on the sample liver plain scan image and the sample liver enhanced image, and a coarse segmentation result of the liver segmentation of the plain scan image, a liver segmentation result of the enhanced image, a key point heat map of the enhanced image, and a blood vessel segmentation result of the enhanced image are output;

[0033] Through the plain scan liver segmentation model to be trained, the sample liver plain scan image, the sample liver enhanced image and the enhanced image liver segmentation result, fine-grained detection is performed on the plain scan image liver segmentation coarse segmentation result, and the plain scan image liver segmentation target result is output.

[0034] In one embodiment, during the coarse-grained detection process, the parameters of the plain scan liver segmentation model to be trained are updated based on the first loss function, the second loss function, the third loss function and the key point loss function;

[0035] Among them, the first loss function is the loss function between the rough segmentation result of the liver segmentation of the plain scan image and the liver segmentation label of the plain scan image, the second loss function is the loss function between the liver segmentation result of the enhanced image and the liver segmentation label of the enhanced image, the third loss function is the loss function between the blood vessel segmentation result of the enhanced image and the blood vessel segmentation label of the enhanced image, and the key point loss function is the regression loss function between the key point heat map of the enhanced image and the sample key point heat map.

[0036] In one embodiment, during the fine-grained detection process, the parameters of the plain scan liver segmentation model to be trained are updated based on the fourth loss function and the similarity loss function;

[0037] Among them, the fourth loss function is a loss function between the plain scan image liver segmentation target result and the plain scan image liver segmentation label, and the similarity loss function is determined based on the enhanced image liver segmentation result and the plain scan image liver segmentation target result.

[0038] In one embodiment, obtaining a liver plain scan image to be analyzed includes:

[0039] The liver is scanned using photon counting CT to obtain the liver plain scan image to be analyzed.

[0040] The present specification provides a liver fat detection device based on image processing, the device comprising:

[0041] A liver image acquisition module, used for acquiring a liver plain scan image to be analyzed;

[0042] A liver segmentation processing module, used for performing liver segmentation processing based on the liver plain scan image to be analyzed to obtain an initial liver segmentation result;

[0043] A key area extraction module, used to extract key areas based on the initial liver segmentation result and determine a target key measurement area;

[0044] The liver fat detection module is used to perform liver fat detection based on the target key measurement area to obtain a fat detection result corresponding to the target key measurement area.

[0045] An embodiment of the present specification provides a medical imaging device, which includes: a memory, and one or more processors communicatively connected to the memory; the memory stores instructions executable by the one or more processors, and the instructions are executed by the one or more processors to enable the one or more processors to implement the steps of the method described in any of the above embodiments.

[0046] The embodiments of this specification provide a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method described in any of the above embodiments are implemented.

[0047] An embodiment of the present specification provides a computer program product, wherein the computer program product includes instructions, and when the instructions are executed by a processor of a computer device, the computer device is enabled to perform the steps of the method described in any one of the above embodiments.

[0048] In the above-mentioned implementation of the specification, first, a plain scan image of the liver to be analyzed is obtained. Then, based on the plain scan image of the liver to be analyzed, liver segmentation processing is performed to obtain an initial liver segmentation result. Then, based on the initial liver segmentation result, key area extraction is performed to determine the target key measurement area. Finally, liver fat detection is performed based on the target key measurement area to obtain the fat detection result corresponding to the target key measurement area. Compared with liver enhancement examination, obtaining a plain scan image of the liver through a plain scan examination of the liver is faster and less expensive, and the radiation dose to the patient is also greatly reduced. Based on the plain scan image of the liver to be analyzed, liver segmentation processing is performed to obtain an initial liver segmentation result, which gets rid of the limitation of relying on liver blood vessels for segmentation on enhanced images, and effectively reduces the influence of blood vessels on the accuracy of analysis and measurement, thereby improving the accuracy of the segmentation results, thereby improving the accuracy of liver fat detection. The above-mentioned implementation method realizes the digitization and intelligence of the processing process through an automated liver analysis process. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flowchart of a liver fat detection method based on image processing provided in an embodiment of this specification;

[0050] Figure 2a A schematic diagram of the process of the liver fat detection method based on image processing provided in the embodiment of this specification;

[0051] Figure 2b Schematic diagram of initial liver segmentation results at different angles provided for embodiments of this specification;

[0052] Figure 2c A schematic diagram of a transverse MPR and its measurement results provided in an embodiment of this specification;

[0053] Figure 2d A schematic diagram of a 3D pseudo-color reconstruction result provided in an embodiment of this specification;

[0054] Figure 2e Schematic diagram of a visual representation of mean fat measurements in a 2D MPR provided for an embodiment of the present specification;

[0055] Figure 2f A schematic diagram of the mean fat measurement value in a 3D interface provided in an embodiment of the present specification;

[0056] Figure 3a A schematic diagram of a process for obtaining an initial key measurement area provided in an embodiment of this specification;

[0057] Figure 3b A schematic diagram of an initial key measurement area obtained by shrinking a segmented profile inward by a preset distance at different angles provided in an embodiment of this specification;

[0058] Figure 3c A schematic diagram of an initial key measurement area obtained by shrinking a segmented profile inward by a preset distance at different angles provided in an embodiment of this specification;

[0059] Figure 4 A schematic diagram of a process for obtaining an initial key measurement area provided in an embodiment of this specification;

[0060] Figure 5a A schematic diagram of a process for obtaining a target key measurement area provided in an embodiment of this specification;

[0061] Figure 5b A schematic diagram of a target key measurement area provided for an embodiment of this specification;

[0062] Figure 5c A schematic diagram of a target key measurement area provided for an embodiment of this specification;

[0063] Figure 6 A schematic diagram of a process for obtaining a liver segmentation model provided in an embodiment of this specification;

[0064] Figure 7 A schematic diagram of a process for obtaining a heat map of key points of a sample provided in an embodiment of this specification;

[0065] Figure 8a A schematic diagram of a process for training a liver segmentation model provided in an embodiment of this specification;

[0066] Figure 8b A flowchart of training a liver segmentation model provided for an embodiment of this specification;

[0067] Fig. 9 A schematic diagram of a liver fat detection device based on image processing provided in an embodiment of this specification. DETAILED DESCRIPTION

[0068] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0069] As the largest gland and solid organ in the human body, the liver is not only the most active metabolic site, but also a site prone to various diseases, such as hepatitis, liver fibrosis, fatty liver, and cirrhosis. Among them, excessive fat accumulation in the liver forms fatty liver, and its incidence rate is increasing. Early intervention can help reverse the development of diseases such as liver fibrosis and fatty liver. Therefore, timely assessment of the degree of development of symptoms is crucial for clinical treatment.

[0070] Liver puncture biopsy is the gold standard for evaluating liver fibrosis and fatty liver, but its promotion is limited by its trauma. In related technologies, the segmentation model combining deep learning and channel attention can calculate the average fat content of each segment, but does not consider the impact of blood vessels on CT values, resulting in errors. Enhanced CT examination can analyze the relationship between key anatomical positions to form liver segments, but patients need to be exposed to radiation and the examination time is long.

[0071] Based on this, the embodiment of this specification provides a liver fat detection method based on image processing. First, a liver plain scan image to be analyzed is obtained. Then, based on the liver plain scan image to be analyzed, liver segmentation processing is performed to obtain an initial liver segmentation result. Then, based on the initial liver segmentation result, key area extraction is performed to determine the target key measurement area. Finally, liver fat detection is performed based on the target key measurement area to obtain the fat detection result corresponding to the target key measurement area. Compared with liver enhancement examination, obtaining liver plain scan images through liver plain scan examination is faster and cheaper, and the radiation dose to the patient is also greatly reduced. Based on the liver plain scan image to be analyzed, liver segmentation processing is performed to obtain the initial liver segmentation result, which gets rid of the limitation of relying on liver blood vessels for segmentation on enhanced images, and effectively reduces the influence of blood vessels on the accuracy of analysis and measurement, thereby improving the accuracy of segmentation results, thereby improving the accuracy of liver fat detection. The above embodiment realizes the digitization and intelligence of the processing process through an automated liver analysis process.

[0072] For example, see Figure 1 , input 3D liver CT plain scan image to the trained liver segmentation model. The liver region is extracted from the 3D liver CT plain scan image by the trained liver segmentation model to obtain the initial liver segmentation result. The liver region of interest is extracted from the initial liver segmentation result, and the fat content of the liver region of interest is measured to obtain the fat detection result corresponding to the liver region of interest. The fat detection result is displayed as a measurement result, and the diagnosis result is output based on the fat detection result.

[0073] This specification provides a liver fat detection method based on image processing. Figure 2a , the liver fat detection method based on image processing may include the following steps:

[0074] S210, obtaining a liver plain scan image to be analyzed.

[0075] S220 , performing liver segmentation processing based on the liver plain scan image to be analyzed to obtain an initial liver segmentation result.

[0076] In some cases, a plain scan of the liver has the advantages of being faster and having less radiation dose to the patient than an enhanced liver scan. A plain scan of the liver can observe the morphological changes of the liver, which is of great significance for the diagnosis of liver composition analysis, liver fibrosis, cirrhosis, and fatty liver. For example, a plain scan image of the liver obtained by computed tomography (CT) and magnetic resonance imaging (MRI) can be obtained without the use of contrast agents. The generation of plain scan images depends on the density differences of liver tissue, thereby being able to provide basic anatomical information about the morphology of the liver. For example, in plain scan images of the liver of patients with fatty liver, a decrease in the density of the liver can usually be observed. This decrease in density is caused by fatty infiltration, indicating an increase in the fat content in hepatocytes, which affects the imaging characteristics of the liver.

[0077] According to Couinaud theory, the liver can be divided into five lobes and eight segments in terms of anatomical structure. Each segment of the liver has different functions, and each segment can be regarded as an independent anatomical structure with the ability to regenerate after resection, which is of great clinical significance for liver resection and transplantation. Therefore, by evaluating each segment individually, the location of the lesion can be more accurately located, thereby effectively evaluating the condition and formulating a more accurate and personalized surgical plan. Therefore, it is necessary to extract the liver segmentation area from the obtained liver plain scan image to obtain more accurate fat detection.

[0078] Specifically, a medical imaging device is used to scan the liver to obtain a liver plain scan image to be analyzed. The liver plain scan image to be analyzed is used as input data of the trained liver segmentation model, and the liver plain scan image to be analyzed is segmented by the trained liver segmentation model to obtain an initial liver segmentation result. For example, please refer to Figure 2b , Figure 2b The initial liver segmentation results at different angles are shown.

[0079] S230: Extract key regions based on the initial liver segmentation result to determine target key measurement regions.

[0080] Specifically, the boundary area of ​​the initial liver segmentation result is easily affected by blood vessels, which in turn affects the measured CT value, thereby affecting the accuracy of liver fat detection. Therefore, in order to improve the accuracy of detection, it is necessary to extract key areas based on the initial liver segmentation results, aiming to determine areas with important detection value and determine the target key measurement area.

[0081] S240: Perform liver fat detection based on the target key measurement area to obtain a fat detection result corresponding to the target key measurement area.

[0082] Specifically, a fat detection algorithm is applied to identify and quantify the fat content of the target key measurement area based on density threshold, tissue distribution pattern or other features. The fat content is analyzed to obtain the fat detection result corresponding to the target key measurement area. For example, the fat content data is analyzed to calculate the mean, median, range and 95% confidence interval of the fat content in the target key measurement area as the fat detection result.

[0083] In some embodiments, liver fat detection can be performed on the target key measurement area by QCT (Quantitative Computed Tomography) to obtain fat detection results corresponding to the target key measurement area. In other embodiments, dual-energy CT can be used to decompose water and fat to obtain fat detection results corresponding to the target key measurement area.

[0084] For example, after obtaining the fat detection results, the fat detection results can be displayed in 2D and 3D. In 2D display mode, you can choose to view the transverse, coronal and sagittal planes, and each window can independently select an area for detailed analysis. The user can select the measurement area as needed, and the fat detection results of the selected area will be automatically calculated. When switching between different levels or positions, the relevant information will be displayed immediately to ensure that the user can obtain the fat detection results in time. For example, please refer to Figure 2c , Figure 2c The transverse MPR and its measurement results are shown. Figure 2c The fat detection results corresponding to the area 202 in the figure include: volume is 7%, average fat content is 13.1, CT mean is 47, and the degree is classified as mild. Figure 2c The fat detection results corresponding to the area 204 in the figure include: volume is 15%, average fat content is 5.2, CT mean is 68, and degree classification is none.

[0085] On the 3D pseudo-color reconstruction interface, the segmented 3D pseudo-color reconstruction results of the liver can be displayed. Figure 2d Each liver segment area is labeled with its corresponding segment name, and each liver segment area will automatically generate the average fat content, CT mean, and degree classification, and the fat content histogram can also be viewed. The transparency of the reconstruction result is adjustable.

[0086] In 2D multi-planar reconstruction (MPR) and 3D interfaces, the pseudo-color of the fat measurement value of each voxel can be further displayed, and multiple staining levels can be refined, with each color corresponding to a different range of fat content. This visual encoding method allows users to intuitively judge the fat content level of each area. See Figure 2e , shows a visual representation of the mean fat content in a 2D MPR. Figure 2f , showing the mean fat content in the 3D interface.

[0087] It should be noted that after the fat detection results are calculated, the system will automatically batch-transmit all data back and intelligently generate a digital diagnosis report. The digital diagnosis report not only contains an overview of the overall fat content, but also a detailed analysis of the fat characteristics of each liver segment and key areas. Specifically, the digital diagnosis report will be subdivided into different liver segments and key areas, and provide their corresponding fat content values, ratings, and possible lesions, so as to more accurately identify lesion areas and assess liver health. Because the digital diagnosis report provides more accurate clinical information, more targeted diagnosis and treatment plans can be formulated.

[0088] In addition, all information in the digital diagnosis report can be tracked online, updated and modified in real time, and automated patient follow-up can be achieved. The system provides one-stop data collection, automated collation and analysis of patient condition information, and realizes automated, digital and intelligent online tracking and management of patients.

[0089] In the above-mentioned implementation, the implementation of this specification provides a liver fat detection method based on image processing. First, a liver plain scan image to be analyzed is obtained. Next, a liver segmentation process is performed based on the liver plain scan image to be analyzed to obtain an initial liver segmentation result. Then, a key area is extracted based on the initial liver segmentation result to determine the target key measurement area. Finally, liver fat detection is performed based on the target key measurement area to obtain a fat detection result corresponding to the target key measurement area. Compared with liver enhancement examination, obtaining a liver plain scan image through a liver plain scan examination is faster and less expensive, and the radiation dose to the patient is also greatly reduced. Based on the liver plain scan image to be analyzed, a liver segmentation process is performed to obtain an initial liver segmentation result, which gets rid of the limitation of relying on liver blood vessels for segmentation on enhanced images, and effectively reduces the influence of blood vessels on the accuracy of analysis and measurement, thereby improving the accuracy of the segmentation results, thereby improving the accuracy of liver fat detection. The above-mentioned implementation realizes the digitization and intelligence of the processing process through an automated liver analysis process.

[0090] In some embodiments, the target key measurement region is determined by: performing key region extraction on the initial liver segmentation result to obtain the target key measurement region.

[0091] Specifically, the boundary area of ​​the initial liver segmentation result is easily affected by blood vessels, which in turn affects the measured CT value, thereby affecting the accuracy of liver fat detection. Therefore, in order to improve the accuracy of detection, the key area of ​​the initial liver segmentation result can be extracted to determine the specific area that needs to be paid attention to or measured, and obtain the target key measurement area.

[0092] In some embodiments, the initial liver segmentation result is subjected to contour extraction by edge detection algorithm or image segmentation technology to obtain a segmented contour. Then, the segmented contour is reduced inward according to a preset distance to obtain a new segmented contour. The new segmented contour and the area covered by it are the target key measurement area.

[0093] In other embodiments, a partition key point is determined in a segmented area of ​​the initial liver segmentation result according to a specific anatomical structure or functional area. The determined partition key point is used as the center point of the segmented area, and the segmented area is grown from the center point according to a preset volume ratio to obtain a target key measurement area corresponding to the segmented area. The above process is repeated to calculate each segmented area of ​​the initial liver segmentation result, thereby obtaining the target key measurement areas corresponding to all segmented areas.

[0094] In the above implementation, key regions are extracted from the initial liver segmentation results to obtain target key measurement regions, which helps to perform more detailed detection of liver fat, thereby improving the ability to diagnose liver diseases.

[0095] In some implementations, the target key measurement region is determined by: receiving a selection operation of the user based on the initial liver segmentation result or the initial key measurement region, and determining the target key measurement region based on the selection operation.

[0096] The initial key measurement area is obtained by extracting the key area of ​​the initial liver segmentation result.

[0097] Specifically, the user can select a specific liver region to be analyzed on the initial liver segmentation result or the initial key measurement region through the interactive interface. According to the user's selection operation, the selected specific liver region is determined as the target key measurement region.

[0098] In the above implementation, a selection operation of the user based on the initial liver segmentation result or the initial key measurement region is received, and the target key measurement region is flexibly determined based on the selection operation.

[0099] In some embodiments, see Figure 3a , extracting key regions from the initial liver segmentation results to obtain initial key measurement regions may include the following steps:

[0100] S310 , performing contour extraction based on the initial liver segmentation result to obtain a segmented contour.

[0101] S320, shrinking the segmented contour inward by a preset distance to eliminate interference from the blood vessel region, thereby obtaining an initial key measurement region.

[0102] Specifically, the initial liver segmentation result is extracted by edge detection algorithm or image segmentation technology to obtain the segmented contour. In order to improve the accuracy of segmentation, the initially obtained segmented contour needs to be optimized. Since the vascular area will interfere with the clear definition of the liver, it is necessary to shrink the segmented contour inward according to a pre-set distance to eliminate the interference of the vascular area and obtain a new segmented contour. The new segmented contour and the area it covers are the required initial key measurement area.

[0103] For example, a segmentation contour is extracted from the initial liver segmentation result. Then, the segmentation contour is shrunk inward by 1 cm to obtain a new segmentation contour. The new segmentation contour and the area covered by it are the initial key measurement area. Figure 3b , Figure 3c , Figure 3b and Figure 3c The initial key measurement areas obtained by shrinking the segmented contour inward by a preset distance at different angles are shown respectively.

[0104] In the above implementation, contour extraction is performed based on the initial liver segmentation result to obtain a segmented contour, which is then shrunk inward by a preset distance to eliminate interference from the vascular area to obtain an initial key measurement area, which helps to perform a more detailed detection of liver fat, thereby improving the ability to diagnose liver disease.

[0105] In some embodiments, see Figure 4 , extracting key regions from the initial liver segmentation results to obtain initial key measurement regions may include the following steps:

[0106] S410: Determine the key points of the partition according to the initial liver segmentation result.

[0107] S420, taking the partition key point as the center, calculating the initial liver segmentation result according to the preset volume ratio to obtain the initial key measurement area.

[0108] Specifically, according to a specific anatomical structure or functional area, a partition key point is determined in a segmented area of ​​the initial liver segmentation result. The determined partition key point is used as the center point of the segmented area, and the segmented area is grown from the center point according to the preset volume ratio to obtain the initial key measurement area corresponding to the segmented area. The above process is repeated to calculate each segmented area of ​​the initial liver segmentation result, so as to obtain the initial key measurement areas corresponding to all segmented areas.

[0109] In some embodiments, the preset volume ratio may be 70%. Based on the center point of a segmented area in the initial liver segmentation result, the segmented area is calculated according to the volume ratio of 70% to obtain the initial key measurement area corresponding to the segmented area.

[0110] In other embodiments, the preset volume ratio can be 70%. A preset radius and step length are preset. The center point of a segmented area in the initial liver segmentation result is used as a reference, and the volume of a sphere bounded by a preset radius is calculated to obtain the initial volume. Subsequently, with the help of the set step length, the radius of the initial volume is gradually expanded, and the volume of the sphere corresponding to the new radius is calculated. After each expansion, the ratio of the current volume to the total volume of the liver is calculated. When the ratio reaches 70%, the expansion is stopped. For example, the width L / 2 of the preset volume is used as the radius and L / 4 is the step length.

[0111] In the above implementation, the partition key points are determined based on the initial liver segmentation results, and the initial liver segmentation results are calculated according to the preset volume ratio with the partition key points as the center to obtain the initial key measurement area, which helps to perform more detailed detection of liver fat, thereby improving the diagnosis ability of liver disease.

[0112] In some embodiments, see Figure 5a , the method may further include the following steps:

[0113] S510: Extract key regions from the initial liver segmentation result to obtain initial key measurement regions.

[0114] S520 , dividing the initial key measurement area into regions to obtain multiple inscribed geometric bodies and use them as target key measurement regions.

[0115] Specifically, from the initial liver segmentation results, key measurement areas are further extracted to obtain a more accurate measurement area, namely the initial key measurement area. The extracted initial key measurement area is refined to obtain more detailed internal structure information or specific geometric bodies to form multiple inscribed geometric bodies. These geometric bodies may be simple geometric shapes (such as spheres, cylinders, etc.) or more complex structures (such as irregular solids), which can better reflect the morphology and characteristics of each area. The multiple inscribed geometric bodies obtained are used as target key measurement areas. The target key measurement area can be used for subsequent quantitative analysis of liver fat detection.

[0116] In some embodiments, based on the initial liver segmentation result, a partition key point is determined, and with the partition key point as the center, the initial liver segmentation result is calculated according to a preset volume ratio to obtain an initial key measurement area.

[0117] In other embodiments, contour extraction is performed based on the initial liver segmentation result to obtain a segmented contour, and the segmented contour is shrunk inward by a preset distance to obtain an initial key measurement area.

[0118] For example, the initial key measurement area can be further subdivided, and an inscribed cube, sphere or polygonal cube can be selected for each key area. Each inscribed area can be further divided into several unit balls or cubes, for example, 4 equal parts, 8 equal parts, or 16 equal parts, and each of the above divided areas can be used as a target key measurement area. Figure 5b , Figure 5b The red sphere and cube in the figure can be the target key measurement area. Figure 5c , Figure 5c Display segmented areas of the liver in different colors, and Figure 5c The red spheres and cubes in the figure can be the target key measurement areas determined within the segmented area.

[0119] In the above implementation, key areas are extracted from the initial liver segmentation results to obtain initial key measurement areas, and the initial key measurement areas are divided into regions to obtain multiple inscribed geometric bodies and use them as target key measurement areas, which helps to perform more detailed detection of liver fat, thereby improving the ability to diagnose liver diseases.

[0120] In some embodiments, see Figure 6 The initial liver segmentation result is obtained by the plain scan liver segmentation model, which is obtained by the following method:

[0121] S610, obtaining a sample liver plain scan image and a sample liver enhanced image.

[0122] The sample liver plain scan image corresponds to a plain scan image liver segmentation label, and the sample liver enhanced image corresponds to an enhanced image liver segmentation label, an enhanced image blood vessel segmentation label and a sample key point heat map.

[0123] Specifically, a liver enhancement scan is performed by intravenous injection of contrast agent to improve the contrast of areas such as blood vessels and tumors, thereby obtaining an enhanced sample liver enhancement image. In some embodiments, a CT plain scan of the liver area is performed by a medical imaging device to obtain a sample liver plain scan image. In other embodiments, when performing a CT scan, the same area is imaged multiple times at different time points to obtain a multi-phase liver plain scan image. Then, based on the sample liver enhancement image, a screening is performed on the multi-phase liver plain scan image, and a phase liver plain scan image that is compatible with the sample liver enhancement image is selected as the sample liver plain scan image. Among them, the sample liver plain scan image and the sample liver enhancement image are both sequences composed of multiple medical images. In addition, the sample liver plain scan image and the sample liver enhancement image can be different imaging modalities from the same patient, or from different patients.

[0124] According to the liver blood vessels (such as the portal vein and its branches, the inferior vena cava, the left, middle and right hepatic veins) and the eight liver segments of the Couinaud standard, the liver segmentation areas and blood vessel areas of the sample liver enhanced image are annotated to obtain masks corresponding to each segment and blood vessel. The annotated liver segments are used as the enhanced image liver segmentation labels corresponding to the sample liver enhanced image, and the annotated blood vessels are used as the enhanced image blood vessel segmentation labels. Based on the enhanced image liver segmentation labels and the enhanced image blood vessel segmentation labels, liver key points are obtained, and then a sample key point heat map is generated based on the liver key points. In some embodiments, the task of annotating the sample liver enhanced image can be completed by an experienced professional doctor.

[0125] According to the eight liver segments of the Couinaud standard, the liver segmentation area of ​​the sample liver plain scan image is annotated to obtain a mask corresponding to each segment. The annotated liver segment is used as the liver segmentation label of the plain scan image corresponding to the sample liver plain scan image. In some embodiments, the task of annotating the sample liver plain scan image can be completed by an experienced professional doctor.

[0126] It should be noted that since both the sample liver plain scan image and the sample liver enhanced image are sequences composed of multiple medical images, it is necessary to annotate each image in the sample liver plain scan image and the sample liver enhanced image respectively to obtain the plain scan image liver segmentation label corresponding to the sample liver plain scan image, the enhanced image liver segmentation label and the enhanced image vascular segmentation label corresponding to the sample liver enhanced image.

[0127] S620, input the sample liver plain scan image and the sample liver enhanced image into the plain scan liver segmentation model to be trained, and output the plain scan image liver segmentation target result, the plain scan image liver segmentation coarse segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map.

[0128] Specifically, a 3D multi-input and multi-output convolutional neural network is built as the plain scan liver segmentation model to be trained. The sample liver plain scan image and the sample liver enhanced image are input into the plain scan liver segmentation model to be trained, and the plain scan image liver segmentation target result, the plain scan image liver segmentation rough segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map are output.

[0129] In some embodiments, before inputting into the plain scan liver segmentation model to be trained, the sample liver plain scan image and the sample liver enhanced image need to be preprocessed (such as denoising, resampling, gray value truncation, normalization) to obtain input data of the plain scan liver segmentation model to be trained. When the sample liver plain scan image and the sample liver enhanced image are from different patients, the sample liver enhanced image can be used to register the sample liver plain scan image to obtain input data of the plain scan liver segmentation model to be trained.

[0130] S630, based on the plain scan image liver segmentation label, enhanced image liver segmentation label, enhanced image blood vessel segmentation label, sample key point heat map, plain scan image liver segmentation target result, plain scan image liver segmentation coarse segmentation result, enhanced image liver segmentation result, enhanced image blood vessel segmentation result and enhanced image key point heat map, update the parameters of the plain scan liver segmentation model to be trained to obtain a trained plain scan liver segmentation model.

[0131] Specifically, the plain scan liver segmentation model to be trained is a multi-task learning framework, which is divided into a main learning task and an auxiliary learning task. The multi-task learning framework improves the utilization efficiency of liver data and reduces the risk of overfitting during the training process, because different tasks can provide a regularization effect on each other. The sample liver plain scan image and the sample liver enhanced image are input into the plain scan liver segmentation model to be trained. The input of the main learning task in the plain scan liver segmentation model to be trained is the sample liver plain scan image, and its goal is to accurately segment the liver. The learning gold standard of this task is the plain scan image liver segmentation label corresponding to the sample liver plain scan image. The input of the auxiliary learning task in the plain scan liver segmentation model to be trained is the sample liver enhanced image, and its learning gold standard is more extensive, including the enhanced image liver segmentation label corresponding to the sample liver enhanced image, the enhanced image vascular segmentation label and the sample key point heat map. These auxiliary learning tasks help to provide additional anatomical structure information, thereby assisting the main learning task to learn more accurately. The auxiliary learning tasks not only help the main learning task learn parameters, but the existence of the auxiliary learning tasks also promotes faster convergence of the main learning task. This interaction enhances the plain scan liver segmentation model's understanding of liver structure. It should be noted that during the training process, the plain scan liver segmentation model automatically selects the auxiliary learning task when the sample liver enhanced image is input, and automatically selects the main learning task when the sample liver plain scan image is input.

[0132] Therefore, based on the plain scan image liver segmentation label, enhanced image liver segmentation label, enhanced image vascular segmentation label, sample key point heat map, plain scan image liver segmentation target result, plain scan image liver segmentation rough segmentation result, enhanced image liver segmentation result, enhanced image vascular segmentation result and enhanced image key point heat map, the loss value of the plain scan liver segmentation model to be trained can be determined, and the parameters of the plain scan liver segmentation model to be trained can be updated based on the model loss value. By analogy, the updated plain scan liver segmentation model continues to be trained, and when the model training stop condition is reached, a trained plain scan liver segmentation model can be obtained. Among them, the model training stop condition can be that the model loss value tends to converge, or that the training rounds reach a preset number of rounds.

[0133] Exemplarily, the plain scan liver segmentation model can adopt a U-shaped structure, including an encoding stage, a decoding stage and an output layer. In multi-task learning, the parameter sharing strategy in the encoding stage and the decoding stage can be hard parameter sharing, soft parameter sharing or partial structural parameter sharing. Too much parameter sharing may lead to negative transfer problems, while too little may lead to failure to effectively utilize the information difference between the sample liver plain scan image and the sample liver enhanced image. Therefore, in the model design, it is necessary to make a choice based on the balance requirements of different tasks. Different decoding stages and output structures can be set for sample liver plain scan images and sample liver enhanced images for differentiated learning.

[0134] In the above implementation, a sample liver plain scan image and a sample liver enhanced image are obtained, and the sample liver plain scan image and the sample liver enhanced image are input into the plain scan liver segmentation model to be trained, and the plain scan image liver segmentation target result, the plain scan image liver segmentation coarse segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map are output. Based on the plain scan image liver segmentation label, the enhanced image liver segmentation label, the enhanced image blood vessel segmentation label, the sample key point heat map, the plain scan image liver segmentation target result, the plain scan image liver segmentation coarse segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map, the parameters of the plain scan liver segmentation model to be trained are updated to obtain a trained plain scan liver segmentation model. By combining the liver plain scan image and the liver enhanced image, the technical effect of the plain scan liver segmentation model on the liver segmentation task is improved to obtain a more accurate liver segmentation area.

[0135] In some embodiments, see Figure 7 , the sample key point heat map is obtained by the following method:

[0136] S710 , performing liver segmentation and blood vessel segmentation processing on the sample liver enhanced image to obtain a liver mask image.

[0137] S720 , extracting blood vessel bifurcation positions and key anatomical positions based on the liver mask image to obtain key point position information.

[0138] S730: Generate a sample key point heat map based on the key point position information.

[0139] The liver mask image includes a liver segmentation mask and a blood vessel segmentation result mask.

[0140] Specifically, in some embodiments, based on the liver segmentation and blood vessel annotation of the sample liver enhanced image, a liver mask image including a liver segmentation mask and a blood vessel segmentation result mask is obtained. In other embodiments, deep learning technology or computer vision algorithm is used to perform liver segmentation and blood vessel segmentation processing on the sample liver enhanced image to obtain a liver mask image including a liver segmentation mask and a blood vessel segmentation result mask. Based on the generated liver mask image, the key point position information is extracted by analyzing the anatomical positions of the vascular bifurcation points and liver partitions. These key points may include different segmented areas of the liver, important vascular intersections, etc. Using the key point position information extracted from the liver mask image, a sample key point heat map is generated. It should be noted that in the process of generating the heat map, the key point position information is used as the center, and the Gaussian distribution is used to attenuate the heat value from the center to the edge to visualize the density and importance of the key point distribution in the liver. Among them, the key point location information may include: inferior vena cava, bifurcation of the right main trunk of portal vein, right hepatic vein, middle hepatic vein, hepatic fissure, bifurcation of the left main trunk of portal vein, left hepatic apex, and falciform ligament.

[0141] Exemplarily, the location of the vascular bifurcation is usually an important boundary for liver segmentation. For example, the bifurcation of the right hepatic vein and the middle hepatic vein can help divide different segmented areas. The bifurcation of the right main trunk of the portal vein and the left main trunk of the portal vein can divide different areas up and down. When determining the location of the vascular bifurcation, the centerline of the vessel is first extracted according to the mask of the vascular segmentation result. Subsequently, a vascular atlas is established using the extracted centerline. By traversing each point on the vascular atlas, the connection characteristics of the vessel can be analyzed. In the vascular atlas, the degree can be the number of edges directly connected to a point. By analyzing the degree of each point on the vascular atlas, the location of the vascular bifurcation can be identified. In the vascular atlas, a degree greater than 2 indicates that the number of vascular variables derived from the point is greater than 2, so the position of the point with a degree greater than 2 is the bifurcation position. The bifurcation position is set as a key point to obtain the key point position information.

[0142] Key anatomical positions can be positions that have a positive impact on liver segmentation. For example, the apex of the left liver marks the end of segment II at the left end of the liver, and the falciform ligament divides the left lobe of the liver into the left medial segment IV and the lateral segments II and III.

[0143] Key point generation for the liver apex position: First, extract the physical center of the liver to provide a benchmark for subsequent contour extraction and segmentation. Then extract the edge contour of the liver. Next, determine the contours of the left and right liver lobes based on the edge contour and physical center of the liver. Then, search for positions on the left and right liver lobe contours where the second-order derivative is zero and the first-order derivative has an opposite sign. Among them, the position on the left liver lobe contour where the second-order derivative is zero and the first-order derivative has an opposite sign is the left liver apex position, and the position on the right liver lobe contour where the second-order derivative is zero and the first-order derivative has an opposite sign is the right liver apex position. Set the left liver apex position and the right liver apex position as key points to obtain key point position information.

[0144] For the key point generation of the falciform ligament: calculate the boundary line of the liver segments II and III, then determine the middle position of the boundary line, and set this position as the key point to obtain the key point position information.

[0145] In the above implementation, liver segmentation and vascular segmentation processing are performed on the sample liver enhanced image to obtain a liver mask image, and the vascular bifurcation positions and key anatomical positions are extracted based on the liver mask image to obtain key point position information. A sample key point heat map is generated based on the key point position information to narrow the area for subsequent detection and improve the accuracy and precision of key point positioning.

[0146] In some embodiments, see Figure 8a , inputting a sample liver plain scan image and a sample liver enhanced image into a plain scan liver segmentation model to be trained, outputting a plain scan image liver segmentation target result, a plain scan image liver segmentation rough segmentation result, an enhanced image liver segmentation result, an enhanced image blood vessel segmentation result and an enhanced image key point heat map, may include the following steps:

[0147] S810. Perform coarse-grained detection on the sample liver plain scan image and the sample liver enhanced image using the plain scan liver segmentation model to be trained, and output the plain scan image liver segmentation coarse segmentation result, enhanced image liver segmentation result, enhanced image key point heat map, and enhanced image blood vessel segmentation result.

[0148] Specifically, the sample liver plain scan image and the sample liver enhanced image are used as the plain scan liver segmentation model to be trained. Figure 8b, use the plain scan liver segmentation model to be trained to perform coarse-grained detection on the sample liver plain scan image and the sample liver enhanced image to identify different segmented areas, vascular areas and key location information of the liver, and output the plain scan image liver segmentation coarse segmentation results, enhanced image liver segmentation results, enhanced image key point heat map and enhanced image vascular segmentation results. These results provide a preliminary reference and basis for subsequent fine-grained detection. It should be noted that although the sample liver enhanced image is subjected to coarse-grained detection, the enhanced image liver segmentation result obtained is a fine segmentation result. The enhanced image liver segmentation result is used to adjust the final output result corresponding to the sample liver plain scan image, that is, the plain scan image liver segmentation target result, during the fine-grained detection process.

[0149] S820, performing fine-grained detection on the plain scan image liver segmentation coarse segmentation result by using the plain scan liver segmentation model to be trained, the sample liver plain scan image, the sample liver enhanced image and the enhanced image liver segmentation result, and outputting the plain scan image liver segmentation target result.

[0150] For details, please continue to see Figure 8b On the basis of coarse-grained detection, the plain scan liver segmentation model to be trained is further utilized, combined with the sample liver plain scan image, the sample liver enhanced image and the enhanced image liver segmentation results, to perform fine-grained detection on the plain scan image liver segmentation coarse segmentation results, more accurately identify the segmented areas of the liver, and output the plain scan image liver segmentation target results.

[0151] In the above implementation, the accuracy of extracting liver segments is improved through two-stage processing, from coarse-grained detection to fine-grained detection.

[0152] In some embodiments, during the coarse-grained detection process, the parameters of the plain scan liver segmentation model to be trained are updated based on the first loss function, the second loss function, the third loss function, and the key point loss function.

[0153] Among them, the first loss function is the loss function between the rough segmentation result of the liver segmentation of the plain scan image and the liver segmentation label of the plain scan image, the second loss function is the loss function between the liver segmentation result of the enhanced image and the liver segmentation label of the enhanced image, the third loss function is the loss function between the blood vessel segmentation result of the enhanced image and the blood vessel segmentation label of the enhanced image, and the key point loss function is the regression loss function between the key point heat map of the enhanced image and the key point heat map of the sample.

[0154] Specifically, during the coarse-grained detection process, multi-task training was performed, covering vascular segmentation, liver segmentation, liver key point extraction on sample liver enhanced images, and liver segmentation tasks on sample liver plain scan images. Since the vascular segmentation task and liver key point extraction are strongly correlated with liver segmentation, the weights are simultaneously back-propagated and updated as a whole task to play a positive role in the liver segmentation task. Based on the difference between the coarse segmentation result of the liver segmentation of the plain scan image and the liver segmentation label of the plain scan image, the loss function between the coarse segmentation result of the liver segmentation of the plain scan image and the liver segmentation label of the plain scan image is determined to be the first loss function. Based on the difference between the liver segmentation result of the enhanced image and the liver segmentation label of the enhanced image, the loss function between the liver segmentation result of the enhanced image and the liver segmentation label of the enhanced image is determined to be the second loss function. Based on the difference between the vascular segmentation result of the enhanced image and the vascular segmentation label of the enhanced image, the loss function between the vascular segmentation result of the enhanced image and the vascular segmentation label of the enhanced image is determined to be the third loss function. Through regression analysis, the difference between the key point heat map of the enhanced image and the key point heat map of the sample is quantified, and the regression loss function between the key point heat map of the enhanced image and the key point heat map of the sample is determined as the key point loss function. The parameters of the plain scan liver segmentation model to be trained are updated based on the first loss function, the second loss function, the third loss function and the key point loss function.

[0155] For example, during the coarse-grained detection process, the loss function is as follows:

[0156]

[0157] Among them, L CTA_Vessel is the third loss function, L CTA_Segment is the second loss function, L NCCT_segment is the first loss function, L Landmark is the key point loss function, For penalty items.

[0158] It should be noted that the first loss function, the second loss function, and the third loss function can adopt Dice loss, cross entropy loss, or a weighted combination thereof. The key point loss function can be an absolute error loss or a mean square error loss. The first loss function, the second loss function, the third loss function, and the key point loss function can be equal coefficient addition or different coefficient weighting. By adjusting the penalty term, the size of the weight can be effectively controlled to cope with the uncertainty in the task.

[0159] In the above implementation, during the coarse-grained detection process, the parameters of the plain scan liver segmentation model to be trained are updated based on the first loss function, the second loss function, the third loss function and the key point loss function to improve the accuracy of liver segmentation.

[0160] In some embodiments, during the fine-grained detection process, the parameters of the plain scan liver segmentation model to be trained are updated based on the fourth loss function and the similarity loss function.

[0161] Among them, the fourth loss function is the loss function between the liver segmentation target result of the plain scan image and the liver segmentation label of the plain scan image, and the similarity loss function is determined based on the liver segmentation result of the enhanced image and the liver segmentation target result of the plain scan image.

[0162] Specifically, during the coarse-grained detection process, the plain scan liver segmentation model to be trained already has the basic ability to identify key point location information, vascular segmentation results and liver segmentation on the sample liver enhanced image. In order to further improve the robustness of liver segmentation on the sample liver plain scan image and reduce segmentation errors. It is necessary to use supervised training in the fine-grained detection process, combining the segmentation tasks on the sample liver enhanced image and the sample liver plain scan image. During the fine-grained detection process, based on the difference between the plain scan image liver segmentation target result and the plain scan image liver segmentation label, the loss function between the plain scan image liver segmentation target result and the plain scan image liver segmentation label is determined as the fourth loss function. The similarity loss function is used to measure the similarity between the enhanced image liver segmentation result and the plain scan image liver segmentation target result. The similarity loss function utilizes the similar liver segmentation morphology in the enhanced image liver segmentation result and the plain scan image liver segmentation target result. The enhanced image liver segmentation result can be used to adjust the plain scan image liver segmentation target result to ensure the similarity of the segmentation results of the two. In some embodiments, based on the difference between the enhanced image liver segmentation result and the plain scan image liver segmentation target result, the loss function between the enhanced image liver segmentation result and the plain scan image liver segmentation target result is determined to be a similarity loss function. In other embodiments, based on the plain scan image liver segmentation label, the enhanced image liver segmentation label, the difference between the enhanced image liver segmentation result and the plain scan image liver segmentation target result, the similarity loss function is determined. The parameters of the plain scan liver segmentation model to be trained are updated based on the fourth loss function and the similarity loss function.

[0163] For example, during fine-grained detection, the loss function is as follows:

[0164] Loss_2=L NCCT +L similarity

[0165] Among them, L NCCT is the fourth loss function, L similarity is the similarity loss function.

[0166] Since deep supervision is used in the training process, the upsampling loss of each stage in the decoding stage needs to be calculated separately. As the weight, L NCCT The announcement is as follows:

[0167]

[0168] Where n is the number of upsampling times, L dec1 is the output of the last upsampling, L decn is the output of the top-level upsampling of the model,

[0169] It should be noted that the similarity loss function may be a Dice loss or a cross entropy loss between the enhanced liver segmentation result and the plain scan liver segmentation result.

[0170] In the above implementation, during the fine-grained detection process, the parameters of the plain scan liver segmentation model to be trained are updated based on the fourth loss function and the similarity loss function to improve the accuracy of liver segmentation.

[0171] In some embodiments, obtaining the plain scan image of the liver to be analyzed includes: scanning the liver using photon counting CT to obtain the plain scan image of the liver to be analyzed.

[0172] Specifically, in order to improve the quality of liver plain scan images, the liver is scanned using photon counting CT to obtain data collected by the photon counting CT, and then the collected data is reconstructed using an image reconstruction algorithm to obtain a liver plain scan image to be analyzed.

[0173] In the above implementation, photon counting CT is used to scan the liver to obtain a more accurate plain scan image of the liver to be analyzed, thereby obtaining a more accurate initial liver segmentation result, so as to improve the fat detection result.

[0174] This specification provides a liver fat detection device 900 based on image processing. Fig. 9 The liver fat detection device 900 based on image processing includes: a liver image acquisition module 910, a liver segmentation processing module 920, a key area extraction module 930, and a liver fat detection module 940.

[0175] A liver image acquisition module 910 is used to acquire a liver plain scan image to be analyzed;

[0176] A liver segmentation processing module 920 is used to perform liver segmentation processing based on the liver plain scan image to be analyzed to obtain an initial liver segmentation result;

[0177] A key region extraction module 930 is used to extract key regions based on the initial liver segmentation result and determine a target key measurement region;

[0178] The liver fat detection module 940 is used to perform liver fat detection based on the target key measurement area to obtain a fat detection result corresponding to the target key measurement area.

[0179] For a detailed description of the liver fat detection device based on image processing, please refer to the description of the liver fat detection method based on image processing above, which will not be repeated here.

[0180] In some embodiments, a medical imaging device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the method steps in the above embodiments when executing the computer program.

[0181] An embodiment of the present specification provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method in any of the above embodiments are implemented.

[0182] One embodiment of the present specification provides a computer program product, which includes instructions. When the instructions are executed by a processor of a computer device, the computer device can perform the steps of the method of any of the above embodiments.

[0183] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or otherwise processing in a suitable manner if necessary, and then stored in a computer memory.

Claims

1. A liver fat detection method based on image processing, characterized in that: The method comprises: Acquire a plain scan image of the liver to be analyzed; Performing liver segmentation processing based on the liver plain scan image to be analyzed to obtain an initial liver segmentation result; Extract key areas based on the initial liver segmentation results to determine target key measurement areas; Liver fat detection is performed based on the target key measurement area to obtain a fat detection result corresponding to the target key measurement area.

2. The method according to claim 1, characterized in that The target key measurement area is determined by: performing key area extraction on the initial liver segmentation result to obtain the target key measurement area; or receiving a selection operation of the user based on the initial liver segmentation result or the initial key measurement area, and determining the target key measurement area based on the selection operation, wherein the initial key measurement area is obtained by performing key area extraction on the initial liver segmentation result.

3. The method according to claim 2, characterized in that The step of extracting key regions from the initial liver segmentation result to obtain the initial key measurement region includes: Perform contour extraction based on the initial liver segmentation result to obtain a segmented contour; The segmented contour is shrunk inward by a preset distance to eliminate interference from the blood vessel region, thereby obtaining the initial key measurement region.

4. The method according to claim 2, characterized in that The step of extracting key regions from the initial liver segmentation result to obtain the initial key measurement region includes: Determining partition key points according to the initial liver segmentation result; The initial liver segmentation result is calculated based on the partition key point as the center according to the preset volume ratio to obtain the initial key measurement area.

5. The method according to claim 1, characterized in that The method further comprises: Extracting key regions from the initial liver segmentation result to obtain initial key measurement regions; The initial key measurement area is divided into regions to obtain a plurality of inscribed geometric bodies and use them as the target key measurement area.

6. The method according to claim 1, characterized in that The initial liver segmentation result is obtained by a plain scan liver segmentation model, and the plain scan liver segmentation model is obtained by the following method: Acquire a sample liver plain scan image and a sample liver enhanced image; wherein the sample liver plain scan image corresponds to a plain scan image liver segmentation label, and the sample liver enhanced image corresponds to an enhanced image liver segmentation label, an enhanced image blood vessel segmentation label, and a sample key point heat map; Inputting the sample liver plain scan image and the sample liver enhanced image into the plain scan liver segmentation model to be trained, and outputting the plain scan image liver segmentation target result, the plain scan image liver segmentation rough segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map; Based on the plain scan image liver segmentation label, the enhanced image liver segmentation label, the enhanced image blood vessel segmentation label, the sample key point heat map, the plain scan image liver segmentation target result, the plain scan image liver segmentation coarse segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map, the parameters of the plain scan liver segmentation model to be trained are updated to obtain a trained plain scan liver segmentation model.

7. The method according to claim 6, characterized in that The sample key point heat map is obtained by the following method: Performing liver segmentation and blood vessel segmentation processing on the sample liver enhanced image to obtain a liver mask image, wherein the liver mask image includes a liver segmentation mask and a blood vessel segmentation result mask; Extracting vascular bifurcation positions and key anatomical positions based on the liver mask image to obtain key point position information; The sample key point heat map is generated based on the key point position information.

8. The method according to claim 6, characterized in that The sample liver plain scan image and the sample liver enhanced image are input into the plain scan liver segmentation model to be trained, and the plain scan image liver segmentation target result, the plain scan image liver segmentation rough segmentation result, the enhanced image liver segmentation result, the enhanced image blood vessel segmentation result and the enhanced image key point heat map are output, including: Using the plain scan liver segmentation model to be trained, coarse-grained detection is performed on the sample liver plain scan image and the sample liver enhanced image, and a coarse segmentation result of the liver segmentation of the plain scan image, a liver segmentation result of the enhanced image, a key point heat map of the enhanced image, and a blood vessel segmentation result of the enhanced image are output; Through the plain scan liver segmentation model to be trained, the sample liver plain scan image, the sample liver enhanced image and the enhanced image liver segmentation result, fine-grained detection is performed on the plain scan image liver segmentation coarse segmentation result, and the plain scan image liver segmentation target result is output.

9. The method according to claim 8, characterized in that In the process of performing the coarse-grained detection, updating the parameters of the plain scan liver segmentation model to be trained based on the first loss function, the second loss function, the third loss function and the key point loss function; Among them, the first loss function is the loss function between the rough segmentation result of the liver segmentation of the plain scan image and the liver segmentation label of the plain scan image, the second loss function is the loss function between the liver segmentation result of the enhanced image and the liver segmentation label of the enhanced image, the third loss function is the loss function between the blood vessel segmentation result of the enhanced image and the blood vessel segmentation label of the enhanced image, and the key point loss function is the regression loss function between the key point heat map of the enhanced image and the sample key point heat map.

10. The method according to claim 8, characterized in that In the process of performing fine-grained detection, updating the parameters of the plain scan liver segmentation model to be trained based on the fourth loss function and the similarity loss function; Among them, the fourth loss function is a loss function between the plain scan image liver segmentation target result and the plain scan image liver segmentation label, and the similarity loss function is determined based on the enhanced image liver segmentation result and the plain scan image liver segmentation target result.

11. The method according to claim 1, characterized in that The step of obtaining a liver plain scan image to be analyzed comprises: The liver is scanned using photon counting CT to obtain the liver plain scan image to be analyzed.

12. A liver fat detection device based on image processing, characterized in that: The device comprises: A liver image acquisition module, used for acquiring a liver plain scan image to be analyzed; A liver segmentation processing module, used for performing liver segmentation processing based on the liver plain scan image to be analyzed to obtain an initial liver segmentation result; A key area extraction module, used to extract key areas based on the initial liver segmentation result and determine a target key measurement area; The liver fat detection module is used to perform liver fat detection based on the target key measurement area to obtain a fat detection result corresponding to the target key measurement area.

13. A medical imaging device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 11 are implemented.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 11 are implemented.

Images