A processing method for tumor images

Abstract

The present invention relates to a method for processing tumor images, and relates to the technical field of medical image processing. The method comprises the following steps: obtaining a tumor image to be segmented; preprocessing the tumor image to be segmented to generate a preprocessed tumor image to be segmented; determining grayscale distribution characteristics of the preprocessed tumor image to be segmented, and then determining a first image segmentation scheme, performing a first image segmentation on the grayscale distribution characteristics of the preprocessed tumor image to be segmented to generate a first image of interest; determining morphological characteristics of the first image of interest; determining a second image segmentation scheme for the first image of interest based on the morphological characteristics of the first image of interest; performing a second image segmentation on the first image of interest using the second image segmentation scheme to generate a second image of interest; and post-processing the second image of interest to generate a target image. The method has the advantages of realizing automatic segmentation of tumor images and improving the efficiency and accuracy of tumor image segmentation.

Description

Technical Field

[0001] The present invention relates to the technical field of temperature control, and in particular to a method for processing tumor images. Background Art

[0002] Image segmentation, a key step in the transition from image processing to image analysis, divides an image into several specific regions of interest with unique properties. With the increasing aging of the population, the number of cancer patients is also gradually increasing. In order to provide effective treatment to cancer patients as soon as possible, it is necessary to accurately characterize the tumor in the affected area. The traditional method is to use CT imaging technology to capture a three-dimensional CT image of the affected area, then manually define the outline of the tumor in the image (this work can also be called "segmentation" of the tumor). Finally, a three-dimensional modeling system is used to create a corresponding three-dimensional model of the tumor based on the depicted tumor outline for reference and use by medical personnel. However, with the increasing requirements for the accuracy and efficiency of tumor three-dimensional models, manual segmentation can no longer meet the corresponding requirements.

[0003] Therefore, it is necessary to provide a processing method for tumor images to realize automatic segmentation of tumor images and improve the efficiency and accuracy of tumor image segmentation. Summary of the Invention

[0004] The present invention provides a method for processing tumor images, comprising: acquiring a tumor image to be segmented; preprocessing the tumor image to be segmented to generate a preprocessed tumor image to be segmented; determining grayscale distribution characteristics of the preprocessed tumor image to be segmented, and determining a first image segmentation scheme based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented; performing a first image segmentation on the grayscale distribution characteristics of the preprocessed tumor image to be segmented based on the first image segmentation scheme to generate a first image of interest; determining morphological characteristics of the first image of interest; determining a second image segmentation scheme for the first image of interest based on the morphological characteristics of the first image of interest; performing a second image segmentation on the first image of interest using the second image segmentation scheme to generate a second image of interest; and post-processing the second image of interest to generate a target image.

[0005] Furthermore, the grayscale distribution characteristics of the preprocessed tumor image to be segmented are determined, and a first image segmentation scheme is determined based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented, including: obtaining multiple sample tumor images and a first image segmentation scheme corresponding to each sample tumor image; determining multiple grayscale distribution characteristic factors; determining multiple key grayscale distribution characteristic factors based on the multiple sample tumor images, the first image segmentation scheme corresponding to each sample tumor image, and the multiple grayscale distribution characteristic factors; determining the grayscale distribution characteristics of the preprocessed tumor image to be segmented based on the multiple key grayscale distribution characteristic factors; and determining the first image segmentation scheme based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented.

[0006] Furthermore, based on multiple sample tumor images, the first image segmentation scheme corresponding to each sample tumor image and the multiple grayscale distribution characteristic factors, multiple key grayscale distribution characteristic factors are determined, including: for each sample tumor image, determining the grayscale distribution characteristic value of each grayscale distribution characteristic factor corresponding to the sample tumor image; for each grayscale distribution characteristic factor, based on the grayscale distribution characteristic values of any two sample tumor images corresponding to the grayscale distribution characteristic factor, calculating the grayscale distribution single factor similarity of any two sample tumor images corresponding to the grayscale distribution characteristic factor; and using a clustering algorithm to calculate the grayscale distribution single factor similarity of any two sample tumor images according to the grayscale distribution characteristic values of any two sample tumor images corresponding to the grayscale distribution characteristic factor. Corresponding to the grayscale distribution single factor similarity of the grayscale distribution characteristic factor, the multiple sample tumor images are classified to determine multiple sample tumor image groups corresponding to the grayscale distribution characteristic factor; based on the first image segmentation scheme corresponding to each sample tumor image included in the sample tumor image group, the group scheme difference value corresponding to the sample tumor image group is calculated; based on the group scheme difference value corresponding to each sample tumor image group, the comprehensive scheme difference value corresponding to the grayscale distribution characteristic factor is calculated; based on the comprehensive scheme difference value corresponding to the grayscale distribution characteristic factor, it is determined whether the grayscale distribution characteristic factor is a key grayscale distribution characteristic factor.

[0007] Furthermore, determining the grayscale distribution characteristic value of each grayscale distribution characteristic factor corresponding to the sample tumor image includes: clustering the pixels included in the sample tumor image according to the grayscale value of each pixel included in the sample tumor image using a clustering algorithm to determine a plurality of pixel groups; and determining the grayscale distribution characteristic value of each grayscale distribution characteristic factor corresponding to the sample tumor image according to the grayscale value of each pixel included in each pixel group.

[0008] Furthermore, based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented, a first image segmentation scheme is determined, including: determining the first image segmentation method based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented, wherein the first image segmentation method is single-threshold segmentation or multi-threshold segmentation; when the first image segmentation method is single-threshold segmentation, determining the target single threshold value based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented; when the first image segmentation method is multi-threshold segmentation, determining the target multi-threshold value based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented, wherein the first image segmentation scheme includes the first image segmentation method and the target single threshold value or the target multi-threshold value.

[0009] Furthermore, a target single threshold is determined based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented, including: grouping multiple sample tumor images according to the first image segmentation scheme corresponding to the multiple sample tumor images to determine a single-threshold segmentation group and a multi-threshold segmentation group; determining single-threshold similar sample tumor images from the single-threshold segmentation group based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the single-threshold segmentation group; and determining the target single threshold through a single-threshold determination model based on the first image segmentation scheme corresponding to the single-threshold similar sample tumor images and the grayscale distribution characteristics of the preprocessed tumor image to be segmented.

[0010] Furthermore, the target multi-threshold value is determined based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented, including: determining a multi-threshold similar sample tumor image from the multi-threshold segmentation group based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the multi-threshold segmentation group; and determining the target multi-threshold value through a multi-threshold determination model based on the first image segmentation scheme corresponding to the multi-threshold similar sample tumor image and the grayscale distribution characteristics of the preprocessed tumor image to be segmented.

[0011] Furthermore, based on the morphological features of the first image of interest, a second image segmentation scheme for the first image of interest is determined, including: determining multiple preset edge shapes based on multiple sample tumor images; calculating edge rule parameters corresponding to the first image of interest based on the morphological features of the first image of interest and the multiple preset edge shapes; and determining the second image segmentation scheme based on the edge rule parameters corresponding to the first image of interest.

[0012] Furthermore, based on the edge rule parameters corresponding to the first image of interest, a second image segmentation scheme is determined, including: establishing multiple image segmentation models based on multiple preset edge shapes; based on the edge rule parameters corresponding to the first image of interest, determining at least one target image segmentation model from the multiple image segmentation models, wherein the second image segmentation scheme includes the at least one target image segmentation model.

[0013] Furthermore, post-processing the second image of interest to generate a target image includes: performing morphological operations and connected region analysis on the second image of interest to generate the target image.

[0014] Compared with the existing technology, the tumor image processing method provided in this specification has at least the following beneficial effects:

[0015] 1. The preprocessing step can remove noise and interference in the image, making the tumor area clearer, which helps doctors observe and diagnose tumors more accurately. Through two image segmentations, based on grayscale distribution characteristics and morphological characteristics respectively, the tumor area can be extracted more accurately, reducing the possibility of misjudgment and missed diagnosis. Accurate tumor segmentation helps to determine the boundaries and scope of the tumor, providing an important basis for the formulation of subsequent treatment plans. During surgical planning, radiotherapy positioning and other treatment processes, accurate segmentation results can help doctors better understand the target area and improve the accuracy and safety of treatment. Automated and intelligent image processing methods can greatly shorten the time of image processing and improve the efficiency of medical work. Doctors can obtain accurate diagnostic results and treatment recommendations more quickly, thereby providing effective treatment to patients in a timely manner;

[0016] 2. By analyzing multiple sample tumor images and their corresponding segmentation schemes, combined with grayscale distribution feature factors, it is possible to more accurately identify grayscale distribution features that have a significant impact on the segmentation results, thereby guiding subsequent image segmentation. Using clustering algorithms and similarity calculations, it is possible to screen out sample images that are most similar to the image to be segmented, thereby ensuring the accuracy and applicability of the segmentation scheme. Based on a large number of sample images for learning and analysis, it can adapt to tumor images of different types and stages and has stronger generalization capabilities. It is not only applicable to the current data set, but also can, to a certain extent, cope with new types and complexities of tumor images that may appear in the future. By determining the key grayscale distribution feature factors, unnecessary calculation and analysis steps can be reduced, thereby improving segmentation efficiency. In addition, the method can automatically select the most suitable segmentation method (single threshold or multi-threshold segmentation), further improving the efficiency and accuracy of segmentation. The entire process from sample learning to segmentation scheme determination is automated, reducing the impact of manual intervention and subjective judgment. This will help promote the intelligent development of tumor image processing technology and improve the efficiency and accuracy of medical diagnosis;

[0017] 3. By introducing multiple preset edge shapes, this method can more accurately capture the tumor edge features in the first image of interest. These preset edge shapes are determined based on multiple sample tumor images and have broad representativeness and applicability. By calculating the edge rule parameters, the segmentation process can be further refined to ensure that the segmentation results are more consistent with the actual morphology of the tumor, thereby improving segmentation accuracy. This method can automatically select the most suitable image segmentation model based on the morphological characteristics of different tumor images. This adaptability enables the method to cope with various complex situations, such as tumors of different sizes, shapes and locations. By establishing multiple image segmentation models and selecting them based on edge rule parameters, the method can ensure that the best segmentation effect is obtained in different situations. By calculating the preset edge shapes and edge rule parameters, the method can reduce unnecessary calculation and analysis steps, thereby speeding up the segmentation speed. In addition, since the method can automatically select the most suitable image segmentation model, there is no need to manually adjust the segmentation parameters, further improving the segmentation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] This specification will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, like numbers represent like structures, wherein:

[0019] Figure 1 is a flow chart of a method for processing tumor images shown in one embodiment of the present application;

[0020] Figure 2 This is a flowchart of determining a first image segmentation solution shown in one embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following briefly introduces the drawings required for describing the embodiments.

[0022] Figure 1 FIG. 1 is a flow chart of a method for processing tumor images shown in an embodiment of the present application. Figure 1 As shown, a method for processing tumor images may include the following steps.

[0023] Step 110: Acquire a tumor image to be segmented.

[0024] Specifically, tumor images to be segmented typically refer to medical images containing tumor regions that require image segmentation. Image segmentation is a key task in computer vision and medical image analysis, aiming to divide an image into multiple regions or objects, each of which corresponds to a specific structure or lesion in the image. In tumor image segmentation, the goal is typically to identify and extract the tumor region for further analysis and evaluation. This can include measuring characteristics of the tumor, such as size, shape, and location, as well as assessing its relationship to surrounding tissues. This information is crucial for formulating treatment plans, monitoring tumor progression, and evaluating treatment efficacy. Tumor images to be segmented come from a variety of medical imaging modalities, such as computed tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET).

[0025] Step 120 : preprocessing the tumor image to be segmented to generate a preprocessed tumor image to be segmented.

[0026] Specifically, preprocessing the tumor image to be segmented may at least include:

[0027] S11. Adjusting the contrast of the tumor image to be segmented to make the difference between the tumor region and surrounding tissue more obvious, such as by histogram equalization, adaptive histogram equalization, or contrast stretching.

[0028] S12. Apply a denoising algorithm (such as median filtering, Gaussian filtering, etc.) to reduce noise in the tumor image to be segmented. Noise may be introduced by factors such as imaging equipment, transmission process, or image compression.

[0029] S13. Scale the pixel values of the tumor image to be segmented to a specific range (e.g., 0 to 1 or -1 to 1). This helps eliminate brightness differences between different images and makes the segmentation algorithm more stable.

[0030] standardization:

[0031] S14. Adjust the mean and standard deviation of the tumor image to be segmented so that it conforms to a normal distribution. Standardization can improve the robustness of the segmentation algorithm to image changes.

[0032] S15. Remove irrelevant areas in the tumor image to be segmented and retain only the area containing the tumor. This can reduce the amount of calculation and improve the accuracy of segmentation.

[0033] S16. Adjust the size of the image to meet the preset requirements.

[0034] Step 130 : determining the grayscale distribution characteristics of the pre-processed tumor image to be segmented, and determining a first image segmentation scheme based on the grayscale distribution characteristics of the pre-processed tumor image to be segmented.

[0035] Figure 2 This is a flow chart showing a method for determining a first image segmentation scheme in an embodiment of the present application. Figure 2 As shown, in some embodiments, step 130 includes:

[0036] obtaining a plurality of sample tumor images and a first image segmentation scheme corresponding to each sample tumor image;

[0037] Determining a plurality of grayscale distribution characteristic factors;

[0038] determining a plurality of key grayscale distribution characteristic factors according to a plurality of sample tumor images, a first image segmentation scheme corresponding to each sample tumor image, and a plurality of grayscale distribution characteristic factors;

[0039] determining the grayscale distribution characteristics of the preprocessed tumor image to be segmented according to a plurality of key grayscale distribution characteristic factors;

[0040] The first image segmentation scheme is determined according to the grayscale distribution characteristics of the preprocessed tumor image to be segmented.

[0041] In some embodiments, multiple key grayscale distribution characteristic factors are determined based on multiple sample tumor images, a first image segmentation scheme corresponding to each sample tumor image, and multiple grayscale distribution characteristic factors, including:

[0042] For each sample tumor image, determining a grayscale distribution characteristic value corresponding to each grayscale distribution characteristic factor of the sample tumor image;

[0043] For each grayscale distribution characteristic factor, the grayscale distribution single-factor similarity of the grayscale distribution characteristic factors corresponding to any two sample tumor images is calculated based on the grayscale distribution characteristic values of the grayscale distribution characteristic factors corresponding to any two sample tumor images. A clustering algorithm (e.g., a K-means clustering algorithm) is used to classify multiple sample tumor images based on the grayscale distribution single-factor similarity of the grayscale distribution characteristic factors corresponding to any two sample tumor images, and multiple sample tumor image groups corresponding to the grayscale distribution characteristic factors are determined. Based on the first image segmentation scheme corresponding to each sample tumor image included in the sample tumor image group, a group scheme difference value corresponding to the sample tumor image group is calculated. Based on the group scheme difference value corresponding to each sample tumor image group, a comprehensive scheme difference value corresponding to the grayscale distribution characteristic factor is calculated. Based on the comprehensive scheme difference value corresponding to the grayscale distribution characteristic factor, it is determined whether the grayscale distribution characteristic factor is a key grayscale distribution characteristic factor.

[0044] In some embodiments, determining the grayscale distribution characteristic value corresponding to each grayscale distribution characteristic factor of the sample tumor image includes:

[0045] Clustering the pixels included in the sample tumor image according to the grayscale value of each pixel included in the sample tumor image using a clustering algorithm to determine a plurality of pixel groups;

[0046] According to the grayscale value of each pixel included in each pixel group, the grayscale distribution characteristic value of the sample tumor image corresponding to each grayscale distribution characteristic factor is determined.

[0047] Specifically, the multiple grayscale distribution characteristic factors may include a pixel group mean standard deviation factor, a pixel group pixel range factor, a pixel group comprehensive fluctuation factor, etc. Among them, the value of the pixel group mean standard deviation factor may be the standard deviation of the grayscale value means of the pixels of the multiple pixel groups. The value of the pixel group pixel range factor may be the difference between the maximum and minimum values of the grayscale value means of the pixels of the multiple pixel groups. The value of the pixel group comprehensive fluctuation factor may be calculated according to the following formula:

[0048]

[0049] Among them, ε is the value of the comprehensive fluctuation factor of the pixel group, σ k is the standard deviation of the grayscale values of the pixels included in the k-th pixel group, K is the total number of pixel groups, G (k,n) is the grayscale value of the nth pixel included in the kth pixel group, N k is the total number of pixels included in the k-th pixel group.

[0050] The grayscale distribution single factor similarity of the grayscale distribution characteristic factors corresponding to any two sample tumor images can be calculated according to the following formula:

[0051]

[0052] Among them, S i is the grayscale distribution single factor similarity of the i-th grayscale distribution characteristic factor of the two sample tumor images, P1 is the preset parameter, P1 is greater than 0, V (1,i) is the value of the grayscale distribution characteristic factor corresponding to a sample tumor image, V (2,i) is the value of the grayscale distribution characteristic factor corresponding to another sample tumor image.

[0053] For any sample tumor image group, when the first image segmentation method corresponding to any two sample tumor images included in the sample tumor image group is the same, for example, both are single threshold segmentation or both are multi-threshold segmentation, then the scheme difference value corresponding to the two sample tumor images is 0; otherwise, the scheme difference value corresponding to the two sample tumor images is 1.

[0054] The group-scheme difference value can be calculated according to the following formula:

[0055]

[0056] Among them, φ k is the group scheme difference value corresponding to the kth sample tumor image group corresponding to the i-th grayscale distribution characteristic factor, ψ (e,f) is the solution difference value corresponding to the e-th sample tumor image and the f-th sample tumor image in the k-th sample tumor image group corresponding to the i-th grayscale distribution characteristic factor, N (i,k) is the total number of sample tumor images included in the kth sample tumor image group corresponding to the i-th grayscale distribution characteristic factor.

[0057] The comprehensive solution difference value can be calculated according to the following formula:

[0058]

[0059] Among them, φ i is the comprehensive scheme difference value corresponding to the i-th grayscale distribution characteristic factor, and N is the total number of sample tumor images.

[0060] The grayscale distribution characteristic factor whose corresponding comprehensive scheme difference value is less than the comprehensive scheme difference value threshold can be used as the key grayscale distribution characteristic factor.

[0061] In some embodiments, determining a first image segmentation scheme based on the grayscale distribution characteristics of the pre-processed tumor image to be segmented includes:

[0062] determining a first image segmentation method according to the grayscale distribution characteristics of the preprocessed tumor image to be segmented, wherein the first image segmentation method is single threshold segmentation or multi-threshold segmentation;

[0063] When the first image segmentation method is single threshold segmentation, the target single threshold is determined according to the grayscale distribution characteristics of the tumor image to be segmented after preprocessing;

[0064] When the first image segmentation method is multi-threshold segmentation, the target multi-threshold is determined according to the grayscale distribution characteristics of the preprocessed tumor image to be segmented, wherein the first image segmentation scheme includes the first image segmentation method and a target single threshold or a target multi-threshold.

[0065] Specifically, the grayscale distribution characteristics of the preprocessed tumor image to be segmented can be extracted from the preprocessed tumor image to be segmented based on multiple key grayscale distribution feature factors, and the grayscale distribution characteristics of the preprocessed tumor image to be segmented include the values corresponding to each key grayscale distribution feature factor.

[0066] The first image segmentation method can be determined by the segmentation method determination model according to the grayscale distribution characteristics of the preprocessed tumor image to be segmented, wherein the segmentation method determination model can be a convolutional neural network model.

[0067] In some embodiments, determining a target single threshold value based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented includes:

[0068] According to the first image segmentation schemes corresponding to the multiple sample tumor images, the multiple sample tumor images are grouped to determine a single-threshold segmentation group and a multi-threshold segmentation group, wherein the first image segmentation method of each sample tumor image in the single-threshold segmentation group is single-threshold segmentation, and the first image segmentation method of each sample tumor image in the multi-threshold segmentation group is multi-threshold segmentation;

[0069] Determining a single-threshold similar sample tumor image from the single-threshold segmentation group based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the single-threshold segmentation group;

[0070] A single threshold determination model is used to determine the target single threshold according to the first image segmentation scheme corresponding to the single threshold similar sample tumor image and the grayscale distribution characteristics of the preprocessed tumor image to be segmented, wherein the single threshold determination model is a convolutional neural network model.

[0071] Specifically, based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the single threshold segmentation group, the grayscale distribution feature similarity between the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the single threshold segmentation group can be calculated, and the sample tumor images whose grayscale distribution feature similarity is greater than the grayscale distribution feature similarity threshold are used as single threshold similar sample tumor images.

[0072] For example, the grayscale distribution feature similarity can be calculated according to the following formula:

[0073]

[0074] Among them, D i is the similarity between the grayscale distribution characteristics of the tumor image to be segmented after preprocessing and the grayscale distribution characteristics of the sample tumor image included in the single threshold segmentation group, P2 is a preset parameter, P2 is greater than 0, V (0,g) is the value of the g-th key grayscale distribution characteristic factor corresponding to the preprocessed tumor image to be segmented, V (i,g) is the value of the g-th key grayscale distribution feature factor corresponding to the i-th sample tumor image included in the single threshold segmentation group, and G is the total number of key grayscale distribution feature factors.

[0075] In some embodiments, determining a target multi-threshold value based on the grayscale distribution characteristics of the pre-processed tumor image to be segmented includes:

[0076] Determine a multi-threshold similar sample tumor image from the multi-threshold segmentation group based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the multi-threshold segmentation group. Specifically, calculate the grayscale distribution feature similarity between the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the multi-threshold segmentation group based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the multi-threshold segmentation group. The method for calculating the grayscale distribution feature similarity between the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the single-threshold segmentation group is similar to that for calculating the grayscale distribution feature similarity between the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the single-threshold segmentation group, and will not be repeated here.

[0077] The target multi-threshold value is determined by a multi-threshold determination model according to the first image segmentation scheme corresponding to the multi-threshold similar sample tumor image and the grayscale distribution characteristics of the pre-processed tumor image to be segmented, wherein the multi-threshold determination model is a convolutional neural network model.

[0078] Step 140 : According to the first image segmentation scheme, perform first image segmentation on the grayscale distribution characteristics of the pre-processed tumor image to be segmented to generate a first image of interest.

[0079] Specifically, for single-threshold segmentation, pixels in the preprocessed tumor image to be segmented whose grayscale values are less than the target single threshold are classified as foreground (or target). Pixels whose grayscale values are greater than or equal to the target single threshold are classified as background. The foreground portion is used as the first image of interest.

[0080] For multi-threshold segmentation, the pre-processed tumor image to be segmented is divided into n+1 regions based on the target multi-threshold. The grayscale values of the pixels in each region are between two adjacent thresholds (either less than the minimum threshold or greater than the maximum threshold). Each region corresponds to a first image of interest.

[0081] Step 150: Determine the morphological features of the first image of interest.

[0082] Specifically, the morphological feature of the first image of interest may include an edge shape of the first image of interest.

[0083] Step 160: Determine a second image segmentation scheme for the first image of interest based on the morphological features of the first image of interest.

[0084] In some embodiments, step 160 specifically includes:

[0085] determining a plurality of preset edge shapes according to a plurality of sample tumor images;

[0086] Calculating edge rule parameters corresponding to the first image of interest based on the morphological features of the first image of interest and a plurality of preset edge shapes, wherein the edge rule parameters corresponding to the first image of interest may include shape similarities between the morphological features of the first image of interest and each of the preset edge shapes;

[0087] A second image segmentation scheme is determined according to edge rule parameters corresponding to the first image of interest.

[0088] In some embodiments, determining a second image segmentation scheme based on edge rule parameters corresponding to the first image of interest includes:

[0089] Based on multiple preset edge shapes, multiple image segmentation models are established, wherein the image segmentation model can be a convolutional neural network model, and each image segmentation model can correspond to a preset edge rule parameter;

[0090] At least one target image segmentation model is determined from a plurality of image segmentation models according to edge rule parameters corresponding to the first image of interest, wherein the second image segmentation scheme includes the at least one target image segmentation model.

[0091] Specifically, the cosine similarity between the edge rule parameters corresponding to the first image of interest and the preset edge rule parameters corresponding to the image segmentation model can be calculated, and the image segmentation model with cosine similarity greater than the cosine similarity threshold can be used as the target image segmentation model.

[0092] Step 170 : Perform a second image segmentation on the first image of interest using a second image segmentation scheme to generate a second image of interest.

[0093] Step 180: Post-process the second image of interest to generate a target image.

[0094] In some embodiments, step 180 specifically includes:

[0095] Perform morphological operations and connected region analysis on the second image of interest to generate a target image.

[0096] Finally, it should be understood that the embodiments described in this specification are intended only to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly described and illustrated in this specification.

Claims

1. A method for processing tumor images, characterized in that: include: Acquire a tumor image to be segmented; Preprocessing the tumor image to be segmented to generate a preprocessed tumor image to be segmented; Determining the grayscale distribution characteristics of the preprocessed tumor image to be segmented, and determining a first image segmentation scheme based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented; performing a first image segmentation on the grayscale distribution characteristics of the pre-processed tumor image to be segmented according to the first image segmentation scheme to generate a first image of interest; determining morphological features of the first image of interest; determining a second image segmentation scheme for the first image of interest based on the morphological features of the first image of interest; performing a second image segmentation on the first image of interest using the second image segmentation scheme to generate a second image of interest; performing post-processing on the second image of interest to generate a target image; Determine the first image segmentation scheme, including: obtaining a plurality of sample tumor images and a first image segmentation scheme corresponding to each sample tumor image; Clustering the pixels included in the sample tumor image according to the grayscale value of each pixel included in the sample tumor image using a clustering algorithm to determine a plurality of pixel groups, and determining a grayscale distribution characteristic value corresponding to each grayscale distribution characteristic factor of the sample tumor image according to the grayscale value of each pixel included in each pixel group, wherein the plurality of grayscale distribution characteristic factors include a pixel group mean standard deviation factor, a pixel group pixel range factor, and a pixel group comprehensive fluctuation factor; According to multiple sample tumor images, the first image segmentation scheme corresponding to each sample tumor image and the multiple grayscale distribution characteristic factors, multiple key grayscale distribution characteristic factors are determined. Specifically, for each sample tumor image, the grayscale distribution characteristic value of each grayscale distribution characteristic factor corresponding to the sample tumor image is determined; for each grayscale distribution characteristic factor, the grayscale distribution single factor similarity of any two sample tumor images corresponding to the grayscale distribution characteristic factor is calculated based on the grayscale distribution characteristic values of any two sample tumor images corresponding to the grayscale distribution characteristic factor, and a clustering algorithm is used to obtain the grayscale distribution single factor similarity of any two sample tumor images according to the grayscale distribution characteristic values of any two sample tumor images corresponding to the grayscale distribution characteristic factor. Classifying the plurality of sample tumor images according to the grayscale distribution single-factor similarity of the grayscale distribution characteristic factor to determine a plurality of sample tumor image groups corresponding to the grayscale distribution characteristic factor; calculating a group-scheme difference value corresponding to the sample tumor image group according to the first image segmentation scheme corresponding to each sample tumor image included in the sample tumor image group; calculating a comprehensive scheme difference value corresponding to the grayscale distribution characteristic factor according to the group-scheme difference value corresponding to each sample tumor image group; and taking a grayscale distribution characteristic factor having a corresponding comprehensive scheme difference value less than a comprehensive scheme difference value threshold as a key grayscale distribution characteristic factor; determining the grayscale distribution characteristics of the preprocessed tumor image to be segmented according to the multiple key grayscale distribution characteristic factors; A first image segmentation scheme is determined according to the grayscale distribution characteristics of the preprocessed tumor image to be segmented, wherein the first image segmentation scheme corresponds to a first image segmentation method of single threshold segmentation or multi-threshold segmentation.

2. The method for processing tumor images according to claim 1, characterized in that: Determining a first image segmentation scheme based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented includes: determining a first image segmentation method according to the grayscale distribution characteristics of the preprocessed tumor image to be segmented; When the first image segmentation method is single threshold segmentation, determining a target single threshold according to the grayscale distribution characteristics of the preprocessed tumor image to be segmented; When the first image segmentation method is multi-threshold segmentation, the target multi-threshold is determined according to the grayscale distribution characteristics of the preprocessed tumor image to be segmented, wherein the first image segmentation scheme includes the first image segmentation method and a target single threshold or a target multi-threshold.

3. The method for processing tumor images according to claim 2, characterized in that: Determining a target single threshold according to the grayscale distribution characteristics of the preprocessed tumor image to be segmented includes: Grouping the multiple sample tumor images according to the first image segmentation scheme corresponding to the multiple sample tumor images to determine a single-threshold segmentation group and a multi-threshold segmentation group; Determining a single-threshold similar sample tumor image from the single-threshold segmentation group based on the grayscale distribution characteristics of the preprocessed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the single-threshold segmentation group; The target single threshold is determined by a single threshold determination model according to a first image segmentation scheme corresponding to the single threshold similar sample tumor image and the grayscale distribution characteristics of the pre-processed tumor image to be segmented.

4. The method for processing tumor images according to claim 3, characterized in that: Determining target multi-threshold values according to the grayscale distribution characteristics of the pre-processed tumor image to be segmented includes: Determining a multi-threshold similar sample tumor image from the multi-threshold segmentation group based on the grayscale distribution characteristics of the pre-processed tumor image to be segmented and the grayscale distribution characteristics of the sample tumor images included in the multi-threshold segmentation group; The target multi-threshold value is determined by a multi-threshold determination model according to a first image segmentation scheme corresponding to the multi-threshold similar sample tumor image and the grayscale distribution characteristics of the pre-processed tumor image to be segmented.

5. A method for processing tumor images according to any one of claims 2 to 4, characterized in that: Determining a second image segmentation scheme for the first image of interest according to the morphological features of the first image of interest includes: determining a plurality of preset edge shapes according to a plurality of sample tumor images; Calculating edge rule parameters corresponding to the first image of interest based on the morphological features of the first image of interest and a plurality of preset edge shapes; A second image segmentation scheme is determined according to edge rule parameters corresponding to the first image of interest.

6. The method for processing tumor images according to claim 5, characterized in that: Determining a second image segmentation scheme according to edge rule parameters corresponding to the first image of interest includes: Based on multiple preset edge shapes, multiple image segmentation models are established; At least one target image segmentation model is determined from the multiple image segmentation models according to edge rule parameters corresponding to the first image of interest, wherein the second image segmentation scheme includes the at least one target image segmentation model.

7. A method for processing tumor images according to any one of claims 1 to 4, characterized in that: Post-processing the second image of interest to generate a target image includes: Performing morphological operations and connected region analysis on the second image of interest to generate the target image.

Images