A method and device for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma

Abstract

The present application belongs to the field of image processing technology, and discloses an analysis method and device for a two-dimensional shear wave elastic image of hepatocellular carcinoma. By collecting a two-dimensional shear wave elastic image, a two-dimensional grayscale image, an elastic grayscale image, and a pixel elastic value matrix in the measurement frame are obtained; an automatic tumor edge segmentation model is established, and the two-dimensional grayscale image to be processed is automatically segmented to obtain the tumor edge, and the tumor edge is migrated to the elastic grayscale image to obtain the tumor edge of the elastic grayscale image; based on the tumor edge of the elastic grayscale image, the intratumoral elastic grayscale image and the adjacent tumor elastic grayscale image are determined, the intratumoral hardness value and the adjacent tumor hardness value are calculated, and the intratumoral feature set and the adjacent tumor feature set are extracted; the intratumoral hardness value, the adjacent tumor hardness value, the intratumoral feature set, and the adjacent tumor feature set are input into the analysis decision system to obtain the analysis result. Deep information mining of the hepatocellular carcinoma shear wave elastic image is realized, and the image utilization rate and analysis and processing efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method and device for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma. Background Art

[0002] Hepatocellular carcinoma is the sixth most common malignant tumor in the world and the fourth most common malignant tumor in terms of mortality. Ultrasound plays an important role in the diagnosis and treatment of hepatocellular carcinoma. The new ultrasound technology, two-dimensional shear wave elastography (2D-SWE), can achieve non-invasive measurement of tissue hardness values. It displays the elasticity map of liver tissue and liver tumors through color coding, and intuitively reflects the hardness distribution of liver tissue and liver tumors. At present, the analysis of two-dimensional shear wave elastography images of hepatocellular carcinoma mainly relies on manual operation. After manually obtaining the shear wave elastography image of hepatocellular carcinoma, the human eye is first required to determine the edge of the tumor in the grayscale image, and measure the hardness inside the tumor and around the cancer separately. Due to the contour characteristics of the tumor, it is difficult to fully reflect the peritumoral hardness by setting a single region of interest (ROI). It is necessary to manually set multiple ROIs along the edge of the tumor for multi-point measurement, and finally the final peritumoral hardness value is obtained by manually calculating the average of multiple hardness values. However, when manually measuring hardness values, in order to ensure the stability of the measurement, a smaller ROI is usually selected during measurement, and the ROI tends to be placed at a relatively uniform position in the image for measurement, thereby reducing the overall utilization rate of shear wave elastic images, and the measurement results are highly subjective; the workload of manual measurement and calculation is huge, and sampling errors inevitably exist between different operators, which is not conducive to the standardization of elastic image data processing; interpreting elastic images only from a single hardness value is somewhat one-sided and cannot fully explore the deep-level data of elastic images. Summary of the invention

[0003] To this end, the embodiments of the present application provide a method and device for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma, so as to realize the mining of deeper information of shear wave elastic images of hepatocellular carcinoma and improve the utilization rate of images.

[0004] In a first aspect, the present application provides a method for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma.

[0005] This application is achieved through the following technical solutions:

[0006] A method for analyzing a two-dimensional shear wave elastic image of hepatocellular carcinoma, the method comprising:

[0007] Collecting a two-dimensional shear wave elastic image of a patient with hepatocellular carcinoma, and obtaining a two-dimensional grayscale image, an elastic grayscale image, and a pixel elastic value matrix corresponding to the two-dimensional shear wave elastic image in a measurement frame;

[0008] Establishing a tumor edge automatic segmentation model, inputting the two-dimensional grayscale image to be processed into the tumor edge automatic segmentation model to obtain the tumor edge of the two-dimensional grayscale image, and migrating the tumor edge of the two-dimensional grayscale image to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image;

[0009] Based on the tumor edge of the elastic grayscale image, segmenting to obtain an intratumoral elastic grayscale image and a paratumoral elastic grayscale image, and calculating an intratumoral hardness value corresponding to the intratumoral elastic grayscale image and a paratumoral hardness value corresponding to the paratumoral elastic grayscale image;

[0010] Determine feature extraction regions in the intratumor elastic grayscale image and the peritumor elastic grayscale image, respectively, and extract an intratumor feature set and a peritumor feature set in the feature extraction regions;

[0011] The intratumoral hardness value, the peritumoral hardness value, the intratumoral feature set and the peritumoral feature set are input into an analysis decision system to obtain an analysis result.

[0012] In a preferred example of the present application, it can be further configured that the step of establishing the tumor edge automatic segmentation model includes:

[0013] The tumor edge in the two-dimensional grayscale image is annotated using annotation software, the annotated two-dimensional grayscale image is input into the nnU-Net network model for training, the Diceloss loss function and the cross entropy loss function are used as the training objective function of the network model, and Adam is used as the optimizer to optimize the network model, and the optimized nnU-Net network model is obtained as the tumor edge automatic segmentation model.

[0014] In a preferred example of the present application, it can be further configured to construct three network models with different configurations, 2DU-net, 3DU-net and cascaded 3D U-net, and select the network model with the best performance as the final network model.

[0015] In a preferred example of the present application, it can be further configured that the step of migrating the tumor edge of the two-dimensional grayscale image to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image includes:

[0016] Utilizing the one-to-one correspondence between the pixels of the two-dimensional grayscale image and the pixels of the elastic grayscale image, the second pixel of the elastic grayscale image is determined according to the first pixel of the tumor edge of the two-dimensional grayscale image, and the tumor edge of the elastic grayscale image is outlined according to the second pixel.

[0017] In a preferred example of the present application, it can be further configured that the step of calculating the intratumoral hardness value corresponding to the intratumoral elastic grayscale image and the peritumoral hardness value corresponding to the peritumoral elastic grayscale image comprises:

[0018] Based on the pixel point elasticity value matrix, an intratumoral elasticity value set and a per-tumoral elasticity value set are determined; based on the intratumoral elasticity value set, an average hardness value corresponding to the intratumoral elastic grayscale image and an average hardness value corresponding to the per-tumoral elastic grayscale image are calculated as the intratumoral hardness value and the per-tumoral hardness value, respectively.

[0019] In a preferred example of the present application, it can be further configured that the step of respectively determining the feature extraction area in the intratumor elastic grayscale image and the adjacent tumor elastic grayscale image comprises:

[0020] Filling the default parts of the intratumoral elastic grayscale image and the paratumoral elastic grayscale image with blank pixels, and standardizing the filled images to the same size;

[0021] The standardized intratumoral elastic grayscale image and the paratumoral elastic grayscale image are intercepted starting with a preset maximum initial interception size, a group of regions of interest are generated based on a preset pixel interval, and it is determined whether there is a candidate region of interest that meets a preset condition in the group of regions of interest;

[0022] If there is no candidate region of interest that meets the preset conditions, the initial interception size is reduced according to the preset step size, and the image is intercepted according to the new interception size until a candidate region of interest that meets the preset conditions is obtained;

[0023] The standard deviation of the pixel elasticity values ​​of the candidate regions of interest is calculated, and the candidate region of interest with the smallest standard deviation is selected as the final feature extraction region.

[0024] In a preferred example of the present application, it can be further configured that the step of extracting the intra-tumor feature set and the peritumoral feature set in the feature extraction area includes:

[0025] First-order statistical features, texture features and wavelet features are extracted in the feature extraction area, key features are selected based on the lasso regression method, and an intra-tumor feature set and a peritumoral feature set are determined.

[0026] In a preferred example of the present application, it can be further configured that the measurement frame includes 1 / 4 of the tumor and 1.0 cm of liver parenchyma around the tumor.

[0027] In a preferred example of the present application, it can be further configured to use 96×96 pixels as the preset maximum initial capture size, and adjust the capture size in steps of 1 pixel or 2 pixels.

[0028] In a second aspect, the present application provides a device for processing a two-dimensional shear wave elastic image of hepatocellular carcinoma, and the present application is implemented by the following technical solutions:

[0029] An analysis device for two-dimensional shear wave elastic images of hepatocellular carcinoma, the device comprising:

[0030] An image processing module is configured to collect a two-dimensional shear wave elastic image of a patient with hepatocellular carcinoma, and obtain a two-dimensional grayscale image, an elastic grayscale image, and a pixel elastic value matrix corresponding to the two-dimensional shear wave elastic image in a measurement frame;

[0031] The edge segmentation module is configured to establish an automatic tumor edge segmentation model, input the two-dimensional grayscale image to be processed into the automatic tumor edge segmentation model to obtain the tumor edge of the two-dimensional grayscale image, and migrate the tumor edge of the two-dimensional grayscale image to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image;

[0032] The data extraction module is configured to segment the tumor edge of the elastic grayscale image to obtain an intratumoral elastic grayscale image and a paratumoral elastic grayscale image, calculate the intratumoral hardness value corresponding to the intratumoral elastic grayscale image and the paratumoral hardness value corresponding to the paratumoral elastic grayscale image; determine feature extraction areas in the intratumoral elastic grayscale image and the paratumoral elastic grayscale image, respectively, and extract an intratumoral feature set and a paratumoral feature set in the feature extraction areas;

[0033] The result output module is configured to input the intratumoral hardness value, the peritumoral hardness value, the intratumoral feature set and the peritumoral feature set into the analysis decision system to obtain the analysis result.

[0034] On the third aspect, the present application is realized through the following technical solutions:

[0035] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above-mentioned methods for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma when executing the computer program.

[0036] In a fourth aspect, the present application provides a computer-readable storage medium.

[0037] This application is achieved through the following technical solutions:

[0038] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of any one of the above-mentioned methods for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma.

[0039] In summary, compared with the prior art, the technical solution provided in the embodiment of the present application has at least the following beneficial effects:

[0040] By collecting two-dimensional shear wave elastic images of patients with hepatocellular carcinoma, a two-dimensional grayscale image, an elastic grayscale image and a pixel elastic value matrix corresponding to the two-dimensional shear wave elastic image in the measurement frame are obtained; an automatic tumor edge segmentation model is established, and the two-dimensional grayscale image to be processed is input into the automatic tumor edge segmentation model to obtain the tumor edge of the two-dimensional grayscale image, and the tumor edge of the two-dimensional grayscale image is migrated to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image; based on the tumor edge of the elastic grayscale image, an intratumoral elastic grayscale image and a paratumoral elastic grayscale image are segmented to obtain the intratumoral hardness value and the paratumoral hardness value are calculated; in the intratumoral elastic grayscale image and the paratumoral elastic grayscale image, feature extraction regions are determined respectively, and an intratumoral feature set and a paratumoral feature set are extracted in the feature extraction region; the intratumoral hardness value, the paratumoral hardness value, the intratumoral feature set and the paratumoral feature set are input into the analysis decision system to obtain the analysis result. It can realize automatic image segmentation and data extraction, solve the problems of large amount of calculation and strong subjectivity caused by manual segmentation and hardness value calculation, realize the mining of deep information of hepatocellular carcinoma shear wave elastic images, and improve image utilization and analysis and processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic flow chart of a method for analyzing a two-dimensional shear wave elastic image of hepatocellular carcinoma provided by an exemplary embodiment of the present application;

[0042] Figure 2 A schematic flow chart of a method for analyzing a two-dimensional shear wave elastic image of hepatocellular carcinoma provided as another exemplary embodiment of the present application;

[0043] Figure 3 A schematic structural diagram of a device for processing two-dimensional shear wave elastic images of hepatocellular carcinoma provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0044] This specific embodiment is merely an explanation of the present application and is not a limitation of the present application. After reading this specification, those skilled in the art may make modifications to the present embodiment without any creative contribution as needed, but such modifications are protected by the patent law as long as they are within the scope of the claims of the present application.

[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] In addition, the term "and / or" in this application is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this application, unless otherwise specified, generally indicates that the associated objects before and after are in an "or" relationship.

[0047] In this application, the terms "first", "second", etc. are used to distinguish identical or similar items with basically the same effects and functions. It should be understood that there is no logical or temporal dependency between "first", "second", and "nth", nor is there any limitation on quantity and execution order.

[0048] The embodiments of the present application are further described in detail below in conjunction with the drawings in the specification.

[0049] In one embodiment of the present application, a method for analyzing a two-dimensional shear wave elastic image of hepatocellular carcinoma is provided, such as Figure 1 As shown, the main steps are described as follows:

[0050] S10: Collect a two-dimensional shear wave elastic image of a patient with hepatocellular carcinoma, and obtain a two-dimensional grayscale image, an elastic grayscale image, and a pixel elastic value matrix corresponding to the two-dimensional shear wave elastic image in a measurement frame.

[0051] Two-dimensional shear wave elastography (2D-SWE) can achieve non-invasive measurement of tissue hardness. The basic principle is that when shear waves propagate in tissues, their propagation speed is closely related to the elasticity or hardness of the tissue. Harder tissues will hinder the propagation of shear waves, causing their speed to decrease, while softer tissues will make shear waves propagate faster. Therefore, by measuring the propagation speed of shear waves in tissues, the elasticity or hardness of tissues can be indirectly evaluated.

[0052] Hepatocellular carcinoma has the biological characteristics of infiltrating and invading the surrounding normal liver tissue, forming micrometastases and microvascular infiltration, which are closely related to the invasiveness and prognosis of hepatocellular carcinoma. Micrometastases and microvascular infiltration cannot be displayed on two-dimensional grayscale ultrasound, but their accumulation in the adjacent liver parenchyma will change the distribution of the shear wave elasticity image of the adjacent liver parenchyma. Secondly, the excessive deposition of extracellular matrix to form adjacent fibrosis is considered to be a mechanical barrier to block the immune infiltration of hepatocellular carcinoma. It is manifested as a high-hardness liver parenchyma surrounding hepatocellular carcinoma on the two-dimensional shear wave elasticity image, that is, the "hard ring sign", which is closely related to the poor prognosis of hepatocellular carcinoma. In addition, malignant tumor tissue usually shows a higher hardness in the tumor than in normal liver tissue due to excessive cell proliferation, rich blood supply, active extracellular matrix remodeling, and excessive deposition of fibrous tissue. Its hardness value has certain value in distinguishing benign and malignant tumors and predicting microvascular infiltration. Therefore, the elastic distribution of the liver parenchyma in and adjacent to hepatocellular carcinoma can provide rich information at the pathological level. Measuring the hardness values ​​of various parts of hepatocellular carcinoma has important clinical significance and application value in evaluating the invasion degree of hepatocellular carcinoma and predicting the recurrence and survival of hepatocellular carcinoma after surgery.

[0053] When collecting preoperative two-dimensional shear wave ultrasound image data of patients with hepatocellular carcinoma, the size of the measurement frame is uniformly set to a rectangle of 4.0cm*3.0cm, and the measurement frame must include 1 / 4 of the tumor and 1.0cm of liver parenchyma around the tumor. Import the original two-dimensional shear wave elastic image in Dicom format, identify the standard color bar in the two-dimensional shear wave elastic image, read the RGB value of each color level in the standard color bar and its corresponding maximum elastic range; read the measurement frame coordinates in the Dicom header file information, and intercept the two-dimensional grayscale image and the two-dimensional shear wave elastic image in the measurement frame from the two-dimensional shear wave elastic image in Dicom format according to the measurement frame coordinates. The two-dimensional shear wave elastic image is a mixture of the surface RGB three-channel color image containing elastic data and the corresponding bottom two-dimensional grayscale image. Therefore, it is necessary to extract the color elastic image from the mixed image and reconstruct the elastic grayscale image based on the hue matching method. Specifically, read the RGB value of each pixel in the color elastic image, match it to the color level of the standard color through the color lookup table, obtain the elastic value of the pixel by conversion, and reconstruct the corresponding elastic grayscale image based on the elastic value matrix of all pixels in the measurement frame. At this point, the two-dimensional grayscale image, elastic grayscale image and pixel elastic value matrix corresponding to the two-dimensional shear wave elastic image in the measurement frame can be obtained.

[0054] S20: Establish an automatic tumor edge segmentation model, input the two-dimensional grayscale image to be processed into the automatic tumor edge segmentation model, obtain the tumor edge of the two-dimensional grayscale image, and migrate the tumor edge of the two-dimensional grayscale image to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image.

[0055] In some embodiments, the process of establishing the tumor edge automatic segmentation model includes: using annotation software to annotate the tumor edge in the two-dimensional grayscale image, inputting the annotated two-dimensional grayscale image into the nnU-Net network model for training, using the Diceloss loss function and the cross entropy loss function as the training objective function of the network model, and using Adam as the optimizer to optimize the network model, and obtaining the optimized nnU-Net network model as the tumor edge automatic segmentation model. Among them, three network models with different configurations are constructed, 2D U-net, 3D U-net and cascaded 3D U-net, and the network model with the best performance is selected as the final network model.

[0056] Specifically, for the two-dimensional grayscale image obtained after processing in step S10, the tumor edge is annotated using Pari image annotation software, and the tumor edge is used as the segmentation gold standard. Through nnU-Net deep learning network analysis, the two-dimensional grayscale images with tumor edge labels are batch-input into the nnU-Net network model for training. The network can be trained by resampling and normalizing the uniform voxel spacing, combining the Diceloss loss function and the cross entropy loss, and using Adam as the optimizer to optimize the network model. During the training process, the learning rate is dynamically adjusted based on the exponential sliding average loss of the training set and the validation set, and the optimized nnU-Net network model is obtained as the tumor edge automatic segmentation model.

[0057] At the same time, in order to limit the overfitting of the model, data augmentation techniques such as random rotation, random scaling, and Gamma adjustment can be applied in real time. The adaptive strategy is used to adjust the network structure, and the post-processing strategy is determined based on the results of the model cross-validation. The nnU-Net network model generates three different U-Net configurations by default: a 2D U-Net, a 3DU-Net, and a cascaded 3D U-Net. The 5-fold cross-validation method is used to verify the segmentation performance of the network model, and the best single model is selected from them. Among them, the 5-fold cross-validation method is to divide the original data set into five subsets of equal size, four of which are used to train the model and one subset is used to verify the model. The above process is repeated 5 times, each time a different subset is selected as the validation set, and the rest are used as the training set. Finally, the performance evaluation results of the 5 times are averaged to obtain the final evaluation result. At the same time, the verification effect can be improved by combining two of the models, so that the integrated model with better performance is used as the final model to establish an automatic segmentation model for the tumor edge. For each case, the standardized two-dimensional grayscale image is input into the above-mentioned tumor edge automatic segmentation model. The model can automatically analyze and integrate quantification to obtain the segmentation of the tumor and adjacent tumors in the two-dimensional grayscale image, that is, the tumor edge inside and adjacent to the tumor. Compared with the time-consuming and labor-intensive analysis method of manually identifying the tumor edge, data processing through artificial intelligence models is more efficient and can reduce subjective errors caused by manual identification.

[0058] After obtaining the tumor edge in the two-dimensional grayscale image, the tumor edge of the two-dimensional grayscale image is migrated to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image, specifically including: using the one-to-one correspondence between the pixel points of the two-dimensional grayscale image and the pixel points of the elastic grayscale image, according to the first pixel point of the tumor edge of the two-dimensional grayscale image, the second pixel point of the elastic grayscale image is determined, and the tumor edge of the elastic grayscale image is outlined according to the second pixel point. Automatic segmentation of the tumor and adjacent tumor in the elastic grayscale image is achieved.

[0059] S30: Based on the tumor edge of the elastic grayscale image, segment the intratumoral elastic grayscale image and the paratumoral elastic grayscale image, and calculate the intratumoral hardness value corresponding to the intratumoral elastic grayscale image and the paratumoral hardness value corresponding to the paratumoral elastic grayscale image.

[0060] Specifically, the intratumoral elastic value set and the paratumoral elastic value set are determined based on the pixel elastic value matrix obtained in step S10, and the average hardness value corresponding to the intratumoral elastic grayscale image and the average hardness value corresponding to the paratumoral elastic grayscale image are calculated based on the intratumoral elastic value set, which are used as the intratumoral hardness value and the paratumoral hardness value, respectively. For example, based on the tumor edge of the elastic grayscale image determined by automatic segmentation, the intratumoral elastic grayscale image and the paratumoral elastic grayscale image can be obtained, and the intratumoral elastic value set corresponding to the intratumoral elastic grayscale image and the paratumoral elastic value set corresponding to the paratumoral elastic grayscale image are read. The original elastic values ​​of all pixels in each set are arranged from small to large, and the actual measurement process is simulated to output the maximum value, minimum value, average value and standard deviation of the hardness value, and the average value is taken as the intratumoral hardness value and the paratumoral hardness value of hepatocellular carcinoma. Compared with the analysis method of performing local measurement by manually delineating the area, the data processing by automatically reading the elastic values ​​of all pixels within the required measurement range is more efficient and can make full use of the information of each pixel in the image.

[0061] S40: Determine feature extraction regions in the intratumor elastic grayscale image and the adjacent tumor elastic grayscale image, respectively, and extract an intratumor feature set and an adjacent tumor feature set in the feature extraction regions.

[0062] In some embodiments, first, the default parts of the intratumoral elastic grayscale image and the per-tumor elastic grayscale image are filled with blank pixels, and the filled images are standardized to the same size; the standardized intratumoral elastic grayscale image and the per-tumor elastic grayscale image are intercepted starting with a preset maximum initial interception size, a group of regions of interest are generated based on a preset pixel interval, and it is determined whether there is a candidate region of interest that meets the preset conditions in the group of regions of interest; if there is no candidate region of interest that meets the preset conditions, the initial interception size is reduced according to a preset step size, and the image is intercepted according to the new interception size until a candidate region of interest that meets the preset conditions is obtained; the standard deviation of the pixel elasticity values ​​of the candidate regions of interest is calculated, and the candidate region of interest with the smallest standard deviation is selected as the final feature extraction region.

[0063] Next, first-order statistical features, texture features, and wavelet features were extracted from the feature extraction area, and key features were selected based on the lasso regression method to determine the intratumoral feature set and the peritumoral feature set. The size of the region of interest was adjusted with a step size of 1 pixel or 2 pixels, starting with 96×96 pixels.

[0064] Take an example to illustrate that since the shape and edge of the tumor are different between different patients, the elastic grayscale image inside the tumor and the elastic grayscale image beside the tumor obtained by the above automatic segmentation need to be standardized so as to extract the imaging genomics features uniformly in the future. First, the blank pixels are used to fill in the missing parts in the elastic grayscale image inside the tumor and the elastic grayscale image beside the tumor, and then the elastic grayscale image inside the tumor and the elastic grayscale image beside the tumor with the blank pixels are adjusted to a uniform size, such as 128×128 pixels, or other sizes greater than 96×96 pixels. A series of different sizes of interception sizes are set to intercept the region of interest, such as 96×96 pixels, 95×95 pixels, 94×94 pixels, etc. First, start with the largest interception size. In this embodiment, 96×96 pixels are used as the initial interception size. The intratumoral elastic grayscale image and the elastic grayscale image beside the tumor are moved at a pixel spacing of 1 pixel to generate a set of regions of interest ROIs with a interception size of 96×96 pixels. 96×96 , detect each region of interest ROI 96×96 Whether there are blank pixels in the image, the region of interest (ROI) containing blank pixels is 96×96 Delete the remaining region of interest ROI that does not contain blank pixels 96×96 As a candidate region of interest ROI 96×96 ; If the previous clipping size (96×96 pixels) fails to obtain a rectangular region of interest that does not contain blank pixels after the entire image, it means that the clipping size is set too large, and the clipping size is selected downward according to the preset step size. For example, if the preset step size is 1 pixel, 95×95 pixels and 94×94 pixels are selected as the clipping size in turn, and the above process is repeated until a candidate region of interest with the largest clipping size and no blank pixels is selected. If the preset step size is 2 pixels, 94×94 pixels and 92×92 pixels are selected as the region of interest size in turn, and the above process is repeated until a candidate region of interest with the largest clipping size and no blank pixels is selected. After determining the candidate region of interest, the standard deviation of the pixel elasticity value in each candidate region of interest is calculated, and the candidate region of interest with the smallest standard deviation of the pixel elasticity value is selected as the final feature extraction region.

[0065] In the case where the rectangular regions of interest have different sizes among different patients, they are standardized to a uniform size during the image preprocessing stage to facilitate subsequent feature extraction. The above region of interest selection method can solve the problem that after image segmentation, the image shapes of the parts of the image are irregular and the pixel sizes are inconsistent among different patients, which makes it impossible to directly extract features. Determining the size of the region of interest in descending order can make the size of the final feature extraction area as large as possible, so that the feature extraction area contains not only homogeneous areas but also non-homogeneous areas, thereby improving the utilization of image information.

[0066] Feature extraction is performed on the elastic grayscale image in the selected region of interest, and the extracted features include first-order statistical features, texture features and wavelet features. Key features are selected based on the lasso regression method, thereby obtaining an intratumoral feature set and a peritumoral feature set. It should be noted that the first-order statistical features are statistical properties based on the elasticity values ​​of the image, such as mean, standard deviation, maximum or minimum value, etc. Texture features describe the local pattern and structure of the image and can reflect the microstructure of the tissue. After extracting a large number of features, in order to reduce the data dimension and reduce the computational complexity, some of the important features are usually selected. In this application, the lasso regression method is used to select key features.

[0067] S50: Input the intratumoral hardness value, the peritumoral hardness value, the intratumoral feature set and the peritumoral feature set into the analysis decision system to obtain the analysis result.

[0068] Based on the intratumoral hardness value, peritumoral hardness value, intratumoral feature set and peritumoral feature set automatically obtained in steps S10-S40, combined with the patient's clinical information and test indicators, the clinical information includes the patient's age, gender, etc., and the test indicators include the patient's laboratory blood test report, etc., a joint model is constructed to build an analysis and decision-making system, and the above indicators are input into the analysis and decision-making system. The analysis and decision-making system can obtain an analysis result of whether the tumor has tumor invasion. The analysis result can be used for clinical auxiliary judgment of whether the tumor is benign or malignant, as well as the possibility of recurrence after resection.

[0069] In another embodiment of the present application, a device for processing a two-dimensional shear wave elastic image of hepatocellular carcinoma is provided. Figure 2 As shown, the device comprises:

[0070] Image processing module 1 is configured to collect a two-dimensional shear wave elastic image of a patient with hepatocellular carcinoma, and obtain a two-dimensional grayscale image, an elastic grayscale image, and a pixel elastic value matrix corresponding to the two-dimensional shear wave elastic image in a measurement frame;

[0071] The edge segmentation module 2 is configured to establish an automatic tumor edge segmentation model, input the two-dimensional grayscale image to be processed into the automatic tumor edge segmentation model to obtain the tumor edge of the two-dimensional grayscale image, and migrate the tumor edge of the two-dimensional grayscale image to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image;

[0072] The data extraction module 3 is configured to segment the tumor edge of the elastic grayscale image to obtain an intratumoral elastic grayscale image and a paratumoral elastic grayscale image, calculate the intratumoral hardness value corresponding to the intratumoral elastic grayscale image and the paratumoral hardness value corresponding to the paratumoral elastic grayscale image; determine feature extraction regions in the intratumoral elastic grayscale image and the paratumoral elastic grayscale image, respectively, and extract an intratumoral feature set and a paratumoral feature set in the feature extraction region;

[0073] The result output module 4 is configured to input the intratumoral hardness value, the peritumoral hardness value, the intratumoral feature set and the peritumoral feature set into the analysis decision system to obtain the analysis result.

[0074] In one embodiment, a computer device is provided, which may be a server.

[0075] The computer device includes a processor, a memory, a network interface and a database connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium has an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, any one of the above-mentioned methods for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma is implemented.

[0076] In one embodiment, a computer-readable storage medium is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-mentioned methods for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma.

[0077] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink), DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0078] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system described in the present application is divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. A method for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma, characterized in that: The method comprises: Collecting a two-dimensional shear wave elastic image of a patient with hepatocellular carcinoma, and obtaining a two-dimensional grayscale image, an elastic grayscale image, and a pixel elastic value matrix corresponding to the two-dimensional shear wave elastic image in a measurement frame; Establishing a tumor edge automatic segmentation model, inputting the two-dimensional grayscale image to be processed into the tumor edge automatic segmentation model to obtain the tumor edge of the two-dimensional grayscale image, and migrating the tumor edge of the two-dimensional grayscale image to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image; Based on the tumor edge of the elastic grayscale image, segment the intratumoral elastic grayscale image and the paratumoral elastic grayscale image, determine the intratumoral elastic value set and the paratumoral elastic value set based on the pixel point elastic value matrix, calculate the average hardness value corresponding to the intratumoral elastic grayscale image based on the intratumoral elastic value set as the intratumoral hardness value, and calculate the average hardness value corresponding to the paratumoral elastic grayscale image based on the paratumoral elastic value set as the paratumoral hardness value; Determine feature extraction regions in the intratumor elastic grayscale image and the peritumor elastic grayscale image, respectively, and extract an intratumor feature set and a peritumor feature set in the feature extraction regions; The intratumoral hardness value, the peritumoral hardness value, the intratumoral feature set and the peritumoral feature set are input into an analysis decision system to obtain an analysis result.

2. The method for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma according to claim 1, characterized in that: The step of establishing the tumor edge automatic segmentation model comprises: The tumor edge in the two-dimensional grayscale image is annotated using annotation software, the annotated two-dimensional grayscale image is input into the nnU-Net network model for training, the Diceloss loss function and the cross entropy loss function are used as the training objective function of the network model, and Adam is used as the optimizer to optimize the network model, and the optimized nnU-Net network model is obtained as the tumor edge automatic segmentation model.

3. The method for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma according to claim 2, characterized in that: Three network models with different configurations are constructed, 2D U-net, 3D U-net and cascaded 3D U-net, and the network model with the best performance is selected as the final network model.

4. The method for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma according to claim 1, characterized in that: The step of migrating the tumor edge of the two-dimensional grayscale image to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image comprises: Utilizing the one-to-one correspondence between the pixels of the two-dimensional grayscale image and the pixels of the elastic grayscale image, the second pixel of the elastic grayscale image is determined according to the first pixel of the tumor edge of the two-dimensional grayscale image, and the tumor edge of the elastic grayscale image is outlined according to the second pixel.

5. The method for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma according to claim 1, characterized in that: The step of respectively determining the feature extraction area in the intratumor elastic grayscale image and the adjacent tumor elastic grayscale image comprises: Filling the default parts of the intratumoral elastic grayscale image and the paratumoral elastic grayscale image with blank pixels, and standardizing the filled images to the same size; The standardized intratumoral elastic grayscale image and the paratumoral elastic grayscale image are intercepted starting with a preset maximum initial interception size, a group of regions of interest are generated based on a preset pixel interval, and it is determined whether there is a candidate region of interest that meets a preset condition in the group of regions of interest; If there is no candidate region of interest that meets the preset conditions, the initial interception size is reduced according to the preset step size, and the image is intercepted according to the new interception size until a candidate region of interest that meets the preset conditions is obtained; The standard deviation of the pixel elasticity values ​​of the candidate regions of interest is calculated, and the candidate region of interest with the smallest standard deviation is selected as the final feature extraction region.

6. The method for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma according to claim 1, characterized in that: The step of extracting the intra-tumor feature set and the peritumoral feature set in the feature extraction region comprises: First-order statistical features, texture features and wavelet features are extracted in the feature extraction area, key features are selected based on the lasso regression method, and an intra-tumor feature set and a peritumoral feature set are determined.

7. The method for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma according to claim 5, characterized in that: The maximum initial capture size is set to 96×96 pixels, and the capture size is adjusted in steps of 1 pixel or 2 pixels.

8. The method for analyzing two-dimensional shear wave elastic images of hepatocellular carcinoma according to claim 1, characterized in that: The measurement frame includes 1 / 4 of the tumor and 1.0 cm of liver parenchyma around the tumor.

9. An analysis device for two-dimensional shear wave elastic images of hepatocellular carcinoma, characterized in that: The device comprises: An image processing module is configured to collect a two-dimensional shear wave elastic image of a patient with hepatocellular carcinoma, and obtain a two-dimensional grayscale image, an elastic grayscale image, and a pixel elastic value matrix corresponding to the two-dimensional shear wave elastic image in a measurement frame; The edge segmentation module is configured to establish an automatic tumor edge segmentation model, input the two-dimensional grayscale image to be processed into the automatic tumor edge segmentation model to obtain the tumor edge of the two-dimensional grayscale image, and migrate the tumor edge of the two-dimensional grayscale image to the corresponding elastic grayscale image to obtain the tumor edge of the elastic grayscale image; The data extraction module is configured to segment the tumor edge of the elastic grayscale image to obtain an intratumoral elastic grayscale image and a paratumoral elastic grayscale image, determine an intratumoral elastic value set and a paratumoral elastic value set based on the pixel elastic value matrix, calculate an average hardness value corresponding to the intratumoral elastic grayscale image based on the intratumoral elastic value set as the intratumoral hardness value, and calculate an average hardness value corresponding to the paratumoral elastic grayscale image based on the paratumoral elastic value set as the paratumoral hardness value; determine feature extraction regions in the intratumoral elastic grayscale image and the paratumoral elastic grayscale image, respectively, and extract an intratumoral feature set and a paratumoral feature set in the feature extraction regions; The result output module is configured to input the intratumoral hardness value, the peritumoral hardness value, the intratumoral feature set and the peritumoral feature set into the analysis decision system to obtain the analysis result.

10. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images