Liver cancer tumor visual detection method based on magnetic resonance image analysis

Abstract

The application relates to a liver cancer tumor visual detection method based on nuclear magnetic resonance image analysis, and belongs to the technical field of image processing.The method comprises the following steps: acquiring a nuclear magnetic resonance image, and extracting a foreground region after pretreatment; acquiring a feature triplet of each pixel point based on the performance of liver cancer tumor in the nuclear magnetic resonance image; determining a first judgment value of each pixel point based on the distribution of the entropy values of a local window taken by the pixel points in the foreground region in different directions, and screening out target pixel points in the foreground region in combination with the feature triplet of each pixel point; determining target pixel points belonging to a liver cancer region based on the matching degree between a clustering cluster where the target pixel points are located and the nuclear magnetic resonance performance of liver cancer, and extracting a liver cancer region from the foreground region based on all the target pixel points belonging to the liver cancer region; and the application screens pixel points according to the difference between the nuclear magnetic resonance performance of liver cancer tumor and normal tissue regions, and accurately segments out a liver cancer region.

Description

Technical Field

[0001] This application relates to the field of image processing technology, specifically to a visual detection method for liver cancer tumors based on nuclear magnetic resonance image analysis. Background Technology

[0002] Magnetic resonance imaging (MRI) is a non-invasive imaging technique that uses magnetic fields and harmless radio waves to generate detailed images of the liver, helping doctors diagnose liver cancer and assess its spread. MRI images allow doctors to observe the size, shape, and internal structure of the liver, as well as any abnormal lumps or tumors.

[0003] Primary liver cancer (HCC) is currently the most common malignant tumor. It is characterized by its insidious onset, high invasiveness, and high malignancy; therefore, early diagnosis is crucial for improving the survival rate of liver cancer patients. Magnetic resonance imaging (MRI) is the primary diagnostic tool and preoperative assessment method for liver cancer. Contrast-enhanced MRI can achieve a detection rate of over 90% for early-stage liver cancer, making it of significant value in the early detection and diagnosis of liver cancer.

[0004] In existing technologies, the determination of the edge of liver cancer region is generally based on the MRI appearance of liver cancer, including tumor morphology and signal intensity. However, when the MRI appearance of liver cancer region in some locations is not significantly different from that of normal tissue, the boundary is relatively blurred. Furthermore, since the location and shape of liver cancer are not fixed, traditional edge detection methods cannot accurately extract the edge of liver cancer region. Summary of the Invention

[0005] This application provides a visual detection method for liver cancer tumors based on magnetic resonance imaging analysis. Based on the characteristic that the left and right hemispheres of the human brain are symmetrical under normal circumstances, and that there are obvious differences in texture between the normal brain tissue on the side where a tumor appears and the other side, the method can accurately identify various types of tumor regions, including gliomas, in the foreground area.

[0006] The visual detection method for liver cancer tumors based on nuclear magnetic resonance image analysis in this application adopts the following technical solution:

[0007] Acquire nuclear magnetic resonance images and extract the foreground region after preprocessing;

[0008] Feature triples for each pixel are obtained based on the appearance of liver cancer tumors in MRI images;

[0009] The first judgment value of each pixel is determined based on the distribution of entropy values ​​of the local window taken by the pixels in the foreground region in different directions. The target pixels in the foreground region are then selected by combining the feature triplet of each pixel.

[0010] Target pixels belonging to the liver cancer region are determined based on the degree of matching between the cluster of the target pixel and the MRI manifestation of liver cancer. Based on all the target pixels belonging to the liver cancer region, the liver cancer region is extracted from the foreground region.

[0011] Preferably, the foreground region is obtained in the following way:

[0012] Image filtering techniques are used to preprocess nuclear magnetic resonance images;

[0013] Connectivity analysis was performed on the filtered nuclear magnetic resonance image to extract several connected components;

[0014] For any connected component, calculate the difference between the maximum and minimum x-coordinates of all pixels in the connected component as the horizontal width of the connected component; calculate the difference between the maximum and minimum y-coordinates of all pixels in the connected component as the vertical width of the connected component.

[0015] The horizontal and vertical widths of all connected regions are counted, and the connected region with the largest horizontal and vertical widths is selected as the foreground region in the MRI image.

[0016] Preferably, the feature triplet for each pixel is obtained as follows:

[0017] A local window of a preset size is obtained with each pixel in the foreground region as the center. The gray-level co-occurrence matrix of the local window in the four directions of 0°, 45°, 90° and 135° is obtained respectively. The direction corresponding to the minimum entropy value of the gray-level co-occurrence matrix in the four directions is taken as the texture direction of each pixel.

[0018] The array consisting of the variance of the entropy values ​​of the gray-level co-occurrence matrices in the four directions, the minimum value of the entropy values ​​of the gray-level co-occurrence matrices in the four directions, and the angle value of the texture direction of each pixel is used as the feature triplet of each pixel.

[0019] Preferably, the first judgment value of each pixel is determined as follows:

[0020] The product of the variance of the entropy values ​​of the gray-level co-occurrence matrices in the four directions obtained from the local window centered on each pixel and the range of the entropy values ​​of the gray-level co-occurrence matrices in the four directions, plus the constant parameter, is used as the denominator.

[0021] The ratio of the minimum entropy value of the gray-level co-occurrence matrix in the four directions to the denominator is used as the first judgment value for each pixel.

[0022] Preferably, the method for filtering out target pixels in the foreground region is as follows:

[0023] The differences between pixels are determined based on the first judgment value of each connected component pixel, the feature triplet of the pixel, and the spatial distribution of the pixel. Then, a clustering algorithm is used to divide the pixels in each connected component into several clusters based on the differences.

[0024] Count the number of clusters corresponding to each connected component, calculate the first quartile of all the counts, and take the pixels in the connected components with a number of clusters greater than the first quartile as the target pixels.

[0025] Preferably, the method for determining the differences between the pixels is as follows:

[0026] For any two non-overlapping pixels within each connected component, calculate the difference between the first judgment threshold between the two pixels, the metric distance between the feature triples of the two pixels, and the Euclidean distance between the two pixels.

[0027] The difference between two pixels is obtained by weighting and fusing all the distances using preset weights.

[0028] Preferably, the method for extracting the liver cancer region from the foreground region is as follows:

[0029] The irregular feature value represented by each target pixel is determined based on the texture direction deviation between each target pixel and the edge pixels of the connected domain in which each target pixel is located.

[0030] The tumor suspicion of each target pixel is determined based on the irregular feature value represented by each target pixel, the irregular feature value represented by the reference point of each target pixel, and the morphological features corresponding to the cluster to which each target pixel belongs.

[0031] The liver cancer region is segmented from the foreground region based on the tumor suspicion of all target pixels within the foreground region.

[0032] Preferably, the irregularity feature value represented by each target pixel is determined as follows:

[0033] Obtain all edge pixels of the connected component containing each target pixel;

[0034] Calculate the absolute value of the difference between the texture direction angle value of each target pixel and any edge pixel, and sum the product of the Euclidean distance between each target pixel and any edge pixel and the absolute value over all edge pixels as the irregularity feature value represented by each target pixel.

[0035] Preferably, the tumor suspicion level of each target pixel is determined as follows:

[0036] The minimum Euclidean distance between each target pixel and all edge pixels in the connected component containing that target pixel is used as the distance threshold.

[0037] Pixels in the cluster where the target pixel is located whose minimum Euclidean distance to the edge pixels of the connected component where the target pixel is located is less than the distance threshold are used as reference points for the target pixel.

[0038] Calculate the sum of the differences between the irregularity feature value represented by each target pixel and the irregularity feature value represented by all reference points;

[0039] The product of the accumulated sum, the variance of the texture direction angle value of the edge pixels of the connected domain where each target pixel is located, and the roundness of the region corresponding to the cluster where each target pixel is located is used as the tumor suspicion of each target pixel.

[0040] Preferably, the method for segmenting the liver cancer region from the foreground region is as follows:

[0041] The tumor suspicion score of each target pixel is calculated, and the threshold of each tumor suspicion score is obtained by using a threshold segmentation algorithm. Target pixels with a tumor suspicion score greater than the threshold are regarded as pixels in the liver cancer region.

[0042] After obtaining all the pixels within the liver cancer region, the edge lines of the liver cancer region are obtained using connected component extraction, thus segmenting the liver cancer region from the foreground region.

[0043] The beneficial effects of this application are as follows: First, image filtering technology is used to preprocess the MRI image to eliminate the influence of the magnetic field generated by the MRI imaging device on the image; second, the foreground region is extracted after preprocessing to improve the efficiency of subsequent segmentation of the liver cancer region edge; then, the feature triplet of each pixel is obtained by analyzing the signal intensity characteristics of the MRI manifestation of liver cancer tumor; then, the first judgment value of each pixel is determined based on the distribution of entropy values ​​of the local window taken by the pixel in the foreground region in different directions, so as to achieve a preliminary assessment of the possibility that the pixel is in the liver cancer region; then, the target pixels in the foreground region are screened by combining the feature triplet of each pixel, and the similarity in the clustering is optimized by considering the severity of different tumor locations and the flatness characteristics of the normal tissue region, thereby improving the accuracy of pixel clustering belonging to the same liver cancer region; then, the target pixels belonging to the liver cancer region are determined based on the matching degree between the cluster where the target pixel is located and the MRI manifestation of liver cancer, and the pixel is screened by using the morphological characteristics of the MRI manifestation of liver cancer tumor, which further improves the accuracy of tumor region segmentation. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating the implementation of the visual detection method for liver cancer tumors based on nuclear magnetic resonance image analysis proposed in this application. Detailed Implementation

[0046] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0047] Examples of the visual detection method for liver cancer tumors based on nuclear magnetic resonance image analysis in this application, such as... Figure 1 As shown, the method includes:

[0048] S001. Acquire nuclear magnetic resonance images and extract the foreground region after preprocessing.

[0049] This application aims to accurately extract the liver cancer tumor region from MRI images by analyzing the edge features of the liver cancer tumor region in MRI images.

[0050] Furthermore, in order to cover as much of the area where liver cancer may be present as possible during magnetic resonance imaging at present, imaging is usually performed within a certain range. This results in the presence of background areas and non-tissue areas in the acquired MRI images. First, it is necessary to screen out the foreground area from the MRI images to improve the efficiency of subsequent detection of the edge of the liver cancer area.

[0051] Specifically, the steps for extracting the foreground region from an MRI image are as follows: First, the MRI image is preprocessed using image filtering techniques to eliminate the influence of the magnetic field generated by the MRI imaging device on the image; second, connected component analysis is performed on the filtered MRI image to extract several connected components; then, for any connected component, the difference between the maximum and minimum x-coordinates of all pixels in each connected component is calculated as the horizontal width of the connected component; the difference between the maximum and minimum y-coordinates of all pixels in each connected component is calculated as the vertical width of the connected component; finally, the horizontal and vertical widths of all the connected components are statistically analyzed, and the connected component with the largest horizontal and vertical widths is selected as the foreground region in the MRI image.

[0052] It should be noted that image filtering is a commonly used technique in the field of image processing. The specific process will not be elaborated here. Commonly used image filtering techniques include, but are not limited to, bilateral filtering denoising, wavelet denoising, and Gaussian filtering denoising. Preferably, in one embodiment of this application, bilateral filtering denoising technique is used for the preprocessing of nuclear magnetic resonance images.

[0053] S002. Extract feature triples for each pixel based on the appearance of liver cancer tumors in MRI images.

[0054] One aspect of liver cancer's appearance on MRI is that the liver cancer tumor typically presents as a low signal intensity on MRI images, while most of the surrounding normal tissue presents as a high signal intensity. Furthermore, the degree of lesion within the liver cancer area is not uniform. Therefore, the texture direction of a pixel can be determined by analyzing the degree of gray-level changes in different directions within a local area.

[0055] In this application, the specific process of obtaining the texture direction is as follows: First, for any pixel in the foreground area, a local window of size N*N is set with each pixel as the center, and the gray-level co-occurrence matrix of the local window in the four directions of 0°, 45°, 90° and 135° is obtained respectively; Second, the variance between the entropy values ​​of the gray-level co-occurrence matrix in the four directions is calculated.

[0056] It should be noted that the purpose of setting a local window is to extract local features of pixels. Therefore, provided that this condition can be met, the size of the local window can be set by the implementer. In this application, the value of N is an odd number between 5 and 11. Preferably, in one embodiment of this application, the size of the local window is 7*7.

[0057] Because the texture of normal tissue in the foreground region often has relatively clear edges and strong local directionality, the gray-level changes at the edges are small. The entropy value of the gray-level co-occurrence matrix obtained along the edge direction is small. However, in other directions, due to the difference in signal intensity of normal tissue in the MRI image, the entropy value of the corresponding gray-level co-occurrence matrix is ​​larger. That is, there will be a minimum value among the entropy values ​​of the four gray-level co-occurrence matrices within the local window defined by the pixels on the texture of normal tissue. The entropy values ​​of the four gray-level co-occurrence matrices within the local window defined by the pixels inside the normal tissue are basically consistent and small. At this time, the variance between the entropy values ​​is also small. Within the liver cancer tumor region, the severity varies at different locations. For example, the brightness of the local areas where the "mother nodule" and "daughter nodule" are located is not consistent in the MRI image. The entropy values ​​of the four gray-level co-occurrence matrices within the local window defined by the pixels in the liver cancer region are all large. Therefore, the entropy values ​​of the gray-level co-occurrence matrices in four different directions can be used to represent the directional characteristics of pixels in the liver cancer region in the MRI image.

[0058] Further, following the above process, four entropy values ​​are obtained for a local window centered on each pixel. The direction corresponding to the minimum of the four entropy values ​​is taken as the texture direction of each pixel. The variance A of the four entropy values ​​of the local window centered on each pixel, and the minimum of the four entropy values ​​are then used as the texture direction. and the angle value of the texture direction of each pixel. The feature triplet that makes up each pixel .

[0059] S003. Based on the distribution of entropy values ​​of local windows taken by pixels in the foreground region in different directions, determine the first judgment value of each pixel, and filter out the target pixels in the foreground region by combining the feature triplet of each pixel.

[0060] For each connected region in the foreground region, the likelihood of the connected region belonging to the liver cancer region is initially determined by the feature triples of the pixels within the connected region.

[0061] Furthermore, considering that the four entropy values ​​calculated for a local window centered on a pixel in the foreground region of normal tissue are relatively close, meaning the texture direction is essentially the same, the texture in such regions lacks strong directionality. Overemphasizing the differences in direction might lead to misjudgment of pixels within the liver cancer region. Therefore, further judgment based on the distribution of the four entropy values ​​is necessary. The above analysis shows that there is a minimum entropy value among the four gray-level co-occurrence matrices within the local window determined by a pixel in the texture of normal tissue; the entropy values ​​of the four gray-level co-occurrence matrices within the local window determined by pixels inside normal tissue are basically consistent and relatively small, with a small variance between the entropy values; and the entropy values ​​of the four gray-level co-occurrence matrices within the local window determined by pixels within the liver cancer region are all relatively large.

[0062] Specifically, for any connected component, taking the i-th connected component as an example, the first discriminant value of each pixel is calculated based on the four entropy values ​​of the local window centered on each pixel in the i-th connected component. This value is used to initially assess the possibility that the pixel is located in the liver cancer region. The formula for calculating the first discriminant value of pixel j is as follows:

[0063] In the formula, It is the first judgment value of pixel j in the i-th connected component. It is the minimum of the four entropy values ​​of the local window centered at pixel j. These are the ranges of the four entropy values, It is a constant parameter used to prevent the variance from being zero. The value can be any positive number not exceeding 0.01. Preferably, in one embodiment of this application, Take the empirical value of 0.001.

[0064] It should be noted that the higher the probability that pixel j belongs to the liver cancer region, the larger the four entropy values ​​of the local window centered on pixel j will be. The larger the value

[0065] Further, following the steps described above, the first judgment value of all pixels within the i-th connected region is obtained. If the i-th connected region belongs to the liver cancer region, the first judgment values ​​of the pixels within the i-th connected region are relatively large for lesion regions of different severity, and there are certain differences between the feature triples of pixels in different lesion regions. When performing clustering, multiple pixel clusters at different locations will be formed. If the i-th connected region belongs to the normal tissue region, the first judgment values ​​of all pixels are relatively small, and only the texture direction of pixels at the edge of the region is inconsistent with the texture direction of pixels inside the region. The texture directions of pixels inside the region are also basically consistent. When performing clustering, a few clusters will be formed.

[0066] Secondly, for any two pixels within the i-th connected region, taking pixel j and pixel j+1 as examples, the difference between the two pixels is calculated based on their NMR spectra in the foreground region. The calculation formula is as follows:

[0067] In the formula, It is the difference between pixel j and pixel j+1. The first preset weight, the second preset weight, and the third preset weight are different and must satisfy the following conditions. , It is the difference between the first judgment value of pixel j and pixel j+1. These are the feature triplet for pixel j and pixel j+1, respectively. yes The distance between them It is the Euclidean distance between pixel j and pixel j+1.

[0068] Furthermore, the difference between any two non-repeating pixels in the i-th connected component is calculated, and a clustering algorithm is used to divide the pixels into several clusters based on the difference. Clustering is a commonly used technique in image processing; the specific process will not be elaborated further. Commonly used clustering algorithms include, but are not limited to, AP (Affinity Propagation) clustering, K-means clustering, Chameleon clustering, and mean-shift clustering. Preferably, in one embodiment of this application, the AP clustering algorithm is used to divide the pixels within the connected component into several clusters.

[0069] It should be understood that The purpose is to measure the difference between the feature triples of pixel j and pixel j+1 during clustering, thereby grouping pixels with similar NMR spectra in the foreground region into a single cluster. Therefore, this measurement is possible. Given the differences between them, different distance metrics can be used in different embodiments, including but not limited to Euclidean distance, DTW distance, and positional variance. Preferably, as an embodiment of this application, for The positional variance between them.

[0070] Further, according to the above steps, the pixels in each connected region of the foreground region are divided into several clusters, the number of clusters corresponding to each connected region is counted, the first quartile of all the said numbers is calculated, and the connected regions with the number of clusters less than or equal to the first quartile are taken as the connected regions corresponding to the normal tissue regions in the foreground region; the connected regions with the number of clusters greater than the first quartile are considered to belong to the liver cancer region, and the pixels in such connected regions are taken as the target pixels.

[0071] S004. Based on the degree of matching between the cluster where the target pixel is located and the MRI manifestation of liver cancer, determine the target pixel belonging to the liver cancer region, and extract the liver cancer region from the foreground region based on all the target pixels belonging to the liver cancer region.

[0072] The MRI findings of liver cancer are also reflected in the fact that the tumor shape is usually round or oval with irregular edges, and necrosis or hemorrhage may even be present inside the tumor. Therefore, the clusters formed by the pixels in the liver cancer area should also appear round or oval. On the other hand, the irregular edges of the tumor area will lead to a greater difference in texture direction between the pixels closer to the edge of the cluster and the texture direction of the other pixels.

[0073] Furthermore, for any target pixel, taking target pixel a as an example, the roundness of the region corresponding to the cluster where target pixel a belongs is calculated to measure whether the region corresponding to the cluster is circular or nearly circular. The roundness determination is a well-known technique in morphology, and the specific process will not be elaborated further.

[0074] Furthermore, the Euclidean distance between each pixel in the cluster where the target pixel a is located and the edge pixels of the connected domain where the target pixel a is located is calculated. The Euclidean distance is used as an evaluation weight to weight the deviation between the texture direction of each pixel and the texture direction of the edge similarity, so as to assess whether there is an irregular edge phenomenon.

[0075] Here, the tumor suspicion score is calculated to characterize the probability that a target pixel belongs to a true liver cancer region. The formula for calculating the tumor suspicion score of target pixel a is:

[0076] In the formula, is the irregularity feature value represented by the target pixel a, n is the number of edge pixels in the connected component where the target pixel a is located, and h represents the h-th edge pixel. It is the Euclidean distance between the target pixel a and the h-th edge pixel. These are the angle values ​​corresponding to the texture direction of the target pixel a and the h-th edge pixel, respectively.

[0077] Furthermore, the minimum Euclidean distance L between the target pixel a and all edge pixels of its connected domain is calculated. Pixels in the cluster containing the target pixel a whose minimum Euclidean distance to the edge pixels of its connected domain is less than L are used as reference points for the target pixel a. Based on the irregularity feature value represented by these reference points, the tumor suspicion level of the target pixel a is determined.

[0078] In the formula, It is the tumor suspicion level of target pixel a. It is an exponential function with the natural constant as its base. It is the distribution variance of the texture direction angle values ​​of the edge pixels of the connected domain where the target pixel a is located. An exponential function is used to avoid the influence of the distribution variance being 0 on the calculation results. is the roundness of the region corresponding to the cluster where target pixel a is located, m is the number of reference points for target pixel a, and b is the b-th reference point for target pixel a. It is the irregularity characteristic value represented by the b-th reference point.

[0079] If the texture direction of a pixel farther from the edge of a connected domain deviates more from that of the edge pixel, it indicates that the texture direction of the edge pixel is significantly inconsistent and has a certain impact on the texture direction of pixels in the connected domain that are closer to the edge pixel. Therefore, the irregularity feature value represented by the reference point in the cluster should be smaller than that represented by the target pixel a compared to the target pixel a.

[0080] Further, following the steps described above, the tumor suspicion level of each target pixel is calculated, and a threshold segmentation algorithm is used to obtain a threshold for each of the tumor suspicion levels. Target pixels with a tumor suspicion level greater than the threshold are identified as pixels within the liver cancer region. After obtaining all pixels within the liver cancer region, the edge lines of the liver cancer region are obtained using connected component extraction, thus segmenting the liver cancer region from the foreground region.

[0081] It should be noted that the purpose of threshold segmentation is to select a threshold from all the tumor suspicion scores, dividing all tumor suspicion scores into two parts. Therefore, provided that the threshold can be obtained, different threshold segmentation algorithms can be selected in different embodiments. Threshold segmentation is a commonly used technique in the field of image processing, including but not limited to Otsu threshold segmentation, global threshold segmentation, and iterative threshold segmentation. Preferably, as an embodiment of this application, the segmentation threshold is obtained using Otsu threshold segmentation.

[0082] It should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

[0083] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A visual detection method for liver cancer tumors based on nuclear magnetic resonance image analysis, characterized in that, include: Acquire nuclear magnetic resonance images and extract the foreground region after preprocessing; Feature triples for each pixel are obtained based on the appearance of liver cancer tumors in MRI images; The first judgment value of each pixel is determined based on the distribution of entropy values ​​of the local window taken by the pixels in the foreground region in different directions. The target pixels in the foreground region are then selected by combining the feature triplet of each pixel. Target pixels belonging to the liver cancer region are determined based on the degree of matching between the cluster where the target pixel belongs and the MRI manifestation of liver cancer. Based on all the target pixels belonging to the liver cancer region, the liver cancer region is extracted from the foreground region. The method for obtaining the feature triplet for each pixel: A local window of a preset size is obtained with each pixel in the foreground region as the center. The gray-level co-occurrence matrix of the local window in the four directions of 0°, 45°, 90° and 135° is obtained respectively. The direction corresponding to the minimum entropy value of the gray-level co-occurrence matrix in the four directions is taken as the texture direction of each pixel. The array consisting of the variance of the entropy values ​​of the gray-level co-occurrence matrices in the four directions, the minimum value of the entropy values ​​of the gray-level co-occurrence matrices in the four directions, and the angle value of the texture direction of each pixel is used as the feature triplet of each pixel. The method for extracting the liver cancer region from the foreground region is as follows: The irregular feature value represented by each target pixel is determined based on the texture direction deviation between each target pixel and the edge pixels of the connected domain in which each target pixel is located. The tumor suspicion of each target pixel is determined based on the irregular feature value represented by each target pixel, the irregular feature value represented by the reference point of each target pixel, and the morphological features corresponding to the cluster to which each target pixel belongs. The liver cancer region is segmented from the foreground region based on the tumor suspicion of all target pixels within the foreground region.

2. The method for visual detection of liver cancer tumors based on nuclear magnetic resonance image analysis according to claim 1, characterized in that, The foreground region is obtained as follows: Image filtering techniques are used to preprocess nuclear magnetic resonance images; Connectivity analysis was performed on the filtered nuclear magnetic resonance image to extract several connected components; For any connected component, calculate the difference between the maximum and minimum x-coordinates of all pixels in the connected component as the horizontal width of the connected component; The difference between the maximum and minimum ordinates of all pixels in each connected component is used as the vertical width of the connected component. The horizontal and vertical widths of all connected regions are counted, and the connected region with the largest horizontal and vertical widths is selected as the foreground region in the MRI image.

3. The method for visual detection of liver cancer tumors based on nuclear magnetic resonance image analysis according to claim 1, characterized in that, The method for determining the first judgment value of each pixel is as follows: The product of the variance of the entropy values ​​of the gray-level co-occurrence matrices in the four directions obtained from the local window centered on each pixel and the range of the entropy values ​​of the gray-level co-occurrence matrices in the four directions, plus the constant parameter, is used as the denominator. The ratio of the minimum entropy value of the gray-level co-occurrence matrix in the four directions to the denominator is used as the first judgment value for each pixel.

4. The method for visual detection of liver cancer tumors based on nuclear magnetic resonance image analysis according to claim 1, characterized in that, The method for filtering out target pixels in the foreground region is as follows: The differences between pixels are determined based on the first judgment value of each connected component pixel, the feature triplet of the pixel, and the spatial distribution of the pixel. Then, a clustering algorithm is used to divide the pixels in each connected component into several clusters based on the differences. Count the number of clusters corresponding to each connected component, calculate the first quartile of all the counts, and take the pixels in the connected components with a number of clusters greater than the first quartile as the target pixels.

5. The method for visual detection of liver cancer tumors based on nuclear magnetic resonance image analysis according to claim 4, characterized in that, The method for determining the differences between the pixels is as follows: For any two non-overlapping pixels within each connected component, calculate the difference in the first judgment value between the two pixels, the metric distance between the feature triples of the two pixels, and the Euclidean distance between the two pixels. The difference between two pixels is obtained by weighting and fusing the difference of the first judgment value, the metric distance and the Euclidean distance using preset weights.

6. The method for visual detection of liver cancer tumors based on nuclear magnetic resonance image analysis according to claim 1, characterized in that, The irregularity feature value represented by each target pixel is determined as follows: Obtain all edge pixels of the connected component containing each target pixel; Calculate the absolute value of the difference between the texture direction angle value of each target pixel and any edge pixel, and sum the product of the Euclidean distance between each target pixel and any edge pixel and the absolute value over all edge pixels as the irregularity feature value represented by each target pixel.

7. The method for visual detection of liver cancer tumors based on nuclear magnetic resonance image analysis according to claim 1, characterized in that, The method for determining the tumor suspicion level of each target pixel is as follows: The minimum Euclidean distance between each target pixel and all edge pixels in the connected component containing that target pixel is used as the distance threshold. Pixels in the cluster where the target pixel is located whose minimum Euclidean distance to the edge pixels of the connected component where the target pixel is located is less than the distance threshold are used as reference points for the target pixel. Calculate the sum of the differences between the irregularity feature value represented by each target pixel and the irregularity feature value represented by all reference points; The product of the accumulated sum, the variance of the texture direction angle value of the edge pixels of the connected domain where each target pixel is located, and the roundness of the region corresponding to the cluster where each target pixel is located is used as the tumor suspicion of each target pixel.

8. The method for visual detection of liver cancer tumors based on nuclear magnetic resonance image analysis according to claim 1, characterized in that, The method for segmenting the liver cancer region from the foreground region is as follows: The tumor suspicion score of each target pixel is calculated, and the threshold of each tumor suspicion score is obtained by using a threshold segmentation algorithm. Target pixels with a tumor suspicion score greater than the threshold are regarded as pixels in the liver cancer region. After obtaining all the pixels within the liver cancer region, the edge lines of the liver cancer region are obtained using connected component extraction, thus segmenting the liver cancer region from the foreground region.

Images