Tumor recognition system based on artificial intelligence and operation method thereof
By designing a tumor recognition system based on artificial intelligence and using image segmentation and feature extraction technology, the problem of low tumor recognition accuracy in the existing technology is solved, and high-accuracy recognition of early and complex tumors is achieved.
Patent Information
- Application Number
- CN202410949824.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2025-01-24
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing tumor recognition technology is less accurate when identifying tumors in the early stages of tumors and complex morphology, and is highly dependent on the quality and labeling accuracy of training data.
A tumor identification system based on artificial intelligence is designed, including an image acquisition module, an image segmentation module, an extraction and labeling module, a fusion comparison module and a recognition module. By splitting the image to be tested into multiple single-channel images, extraction and labeling Features, and classify and identify the feature areas in the composite image through a classification and recognition algorithm.
It improves the ability to extract subtle features from digital pathology slices, enhances the accuracy of tumor identification, and reduces dependence on training data quality and annotation accuracy.
Smart Images
Figure CN119360377A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of tumor recognition, and specifically is a tumor recognition system based on artificial intelligence and an operation method thereof. Background Art
[0002] Tumors are neoplasms formed when a cell in a local tissue loses normal regulation of its growth at the genetic level under the influence of various carcinogenic factors, leading to its clonal abnormal proliferation. The diagnosis of tumors is mainly based on pathological section analysis, but manual reading of pathological sections is a very complex and time-consuming task. In order to help pathologists read pathological sections more quickly and effectively, researchers have proposed digital pathological sections (i.e., converting traditional materialized glass sections into digital pathological sections) and used various algorithms to predict tumor cell areas in digital pathological sections.
[0003] However, there are many types of tumors, and their biological characteristics and pathological manifestations vary. In particular, the symptoms of tumors in the early stages are often not obvious, or the characteristics are slightly different, making it difficult to distinguish them from normal tissues. Existing image recognition technology for tumor recognition is more dependent on the quality of training data and the accuracy of annotations, and the recognition effect is relatively limited for some tumors with more complex shapes and sizes. Therefore, there is an urgent need for an artificial intelligence-based tumor recognition system and its operation method, which can more effectively extract features from digital pathology sections and improve the accuracy of tumor recognition. Summary of the invention
[0004] In order to solve the above problems, the purpose of the present invention is to provide an artificial intelligence-based tumor recognition system and an operation method thereof, which can improve the extraction of subtle features in digital pathological sections, thereby improving the accuracy of tumor recognition.
[0005] In order to achieve the above object, the technical solution of the present invention is as follows:
[0006] A tumor recognition system based on artificial intelligence, comprising: an image acquisition module, an image segmentation module, a label extraction module, a fusion comparison module and a recognition module;
[0007] The image acquisition module is used to obtain the image to be tested;
[0008] The image segmentation module is used to separate the image to be tested into a binary image and a primary color image;
[0009] The extraction and marking module is used to extract and mark the features in the peeled image respectively;
[0010] The fusion comparison module is used to overlap and fuse the binary image and the primary color image to obtain a composite image;
[0011] The recognition module is used to classify and recognize the feature areas in the composite image through a classification recognition algorithm.
[0012] Furthermore, it also includes a correction module; the correction module is used to compare the feature area in the composite image with the feature area in the image to be tested, and determine whether the feature area in the composite image is within the feature area in the image to be tested. If the feature area in the composite image is within the feature area in the image to be tested, the composite image is output to the recognition module; if the feature area in the composite image is outside the feature area in the image to be tested, the features in the peeled image are re-extracted and marked until the feature area in the composite image is within the feature area in the image to be tested.
[0013] Furthermore, it also includes a pre-processing module; the pre-processing module is used to perform denoising on the image to be tested.
[0014] Furthermore, the features include texture features, shape features, and texture features.
[0015] Furthermore, the fusion comparison module is also used to identify the layer in the composite image that is outside the feature area of the image to be tested, and re-extract features of the layer separately; the layer is the layer where the binary image or the original color image is located.
[0016] Furthermore, it also includes an amplification module; the amplification module is used to receive the binary image and the original color image whose features need to be re-extracted, and amplify them according to a preset gradient speed, and send the amplified binary image and the original color image to the extraction mark module.
[0017] Furthermore, the fusion comparison module is also used to divide the fused feature area into overlapping areas and non-overlapping areas, and deepen the edges and contours of the overlapping areas.
[0018] Furthermore, the preset gradient speeds are 40 times, 100 times, 200 times and 400 times.
[0019] Furthermore, the fusion comparison module is also used to overlap the binary image and the original color image according to the coordinates of each pixel on the image to be tested.
[0020] Furthermore, classification and recognition algorithms include neural networks, decision trees, naive Bayes, and support vector machines.
[0021] Further, the primary color image includes an R channel image, a G channel image, and a B channel image.
[0022] An operation method for tumor identification based on artificial intelligence comprises the following steps:
[0023] Step 1: Get the image to be tested;
[0024] Step 2: split the image to be tested into an R channel image, a G channel image, a B channel image and a binary image;
[0025] Step 3: extract and identify the features of the R channel image, the G channel image, the B channel image, the binary image and the image to be tested, and mark the feature areas respectively;
[0026] Step 4: Overlap and compare the marked R channel image, G channel image, B channel image and binary image according to the pixel coordinates in the image to be tested to obtain a fused composite image;
[0027] Step 5: Classify and identify the composite image.
[0028] The above scheme has the following beneficial effects:
[0029] 1. Compared with traditional image recognition technology, this solution divides the image to be tested, i.e., the original image, into R channel image, G channel image, B channel image and binary image, and extracts and marks the features of the tumor in each single channel image to facilitate the extraction of complete features in the later stage.
[0030] 2. Compared with traditional image recognition technology, this solution can highlight the importance of different features by extracting the features of the tumor in each single channel and deepening the edges and contours of the overlapping areas of the features of each single channel.
[0031] 3. Compared with traditional image recognition technology, this solution improves the extraction of subtle features by analyzing the original image layer by layer, making the features clearer and thus improving the accuracy of the recognition results. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 This is a system block diagram of embodiment 1 of the present invention.
[0033] Figure 2 This is a schematic diagram of the judgment logic of Example 1 of the present invention.
[0034] Figure 3 This is a process step diagram of Example 2 of the present invention. DETAILED DESCRIPTION
[0035] The following is further described in detail through specific implementation methods:
[0036] Example 1
[0037] Basically as attached Figure 1-Figure 2 As shown:
[0038] A tumor recognition system based on artificial intelligence includes: an image acquisition module, an image segmentation module, a label extraction module, a fusion comparison module and a recognition module.
[0039] The image acquisition module is used to obtain the image to be tested.
[0040] Specifically, the image to be tested is a color image in JPG format, that is, a high-resolution digital image representing the entire tissue slice is obtained from a glass slide on a computer monitor through WSI technology. After the entire slice is pathologically scanned, the color image is collected through an image acquisition module.
[0041] It also includes a preprocessing module; the preprocessing module is used to perform denoising on the image to be tested.
[0042] The image segmentation module is used to separate the denoised image into a binary image and a primary color image.
[0043] Specifically, the image segmentation module reads the image to be tested through the imread function, and divides the color image into three single-channel images, namely, original color images, specifically, R single-channel image, G single-channel image and B single-channel image, through the split function, and at the same time performs image binarization processing on the image to be tested to obtain a binary image.
[0044] The extraction and marking module is used to extract and mark the features in the peeled image, such as the texture features, shape features and texture features of the tumor, and also performs feature extraction and marking on the image to be tested (i.e. before peeling).
[0045] Specifically, the extraction and marking module performs separate feature extraction on the R single-channel image, the G single-channel image, and the B single-channel image, and marks and divides the feature areas in each image.
[0046] The fusion comparison module is used to overlap and fuse the binary image and the original color image according to the coordinates of each pixel on the image to be tested to obtain a composite image, and at the same time divide the feature area into overlapping areas and non-overlapping areas, and deepen the edges and contours of the overlapping areas.
[0047] The recognition module is used to classify and recognize the feature areas in the composite image through a classification recognition algorithm. The classification recognition algorithms include neural network, decision number, naive Bayes, and support vector machine. The recognition module selects one of the algorithms to classify and recognize the feature areas.
[0048] It also includes a correction module; the correction module is used to compare the feature area in the composite image with the feature area in the image to be tested, and determine whether the feature area in the composite image is within the feature area in the image to be tested. If the feature area in the composite image is within the feature area in the image to be tested, the composite image is output to the recognition module; if the feature area in the composite image is outside the feature area in the image to be tested, the features in the peeled image are re-extracted and marked until the feature area in the composite image is within the feature area in the image to be tested.
[0049] Specifically, if not all of the feature regions in the composite image are within the feature region of the image to be tested, the correction module will re-extract and mark the features of the image that is not within the feature region of the image to be tested.
[0050] It also includes an amplification module, which is used to receive the binary image and the original color image whose features need to be re-extracted, amplify them according to a preset gradient speed, and send the amplified binary image and the original color image to the extraction mark module.
[0051] Specifically, the magnification module gradually increases according to the magnification speed of 40 times, 100 times, 200 times and 400 times, and after each magnification speed, the image (R single-channel image, G single-channel image, B single-channel image and binarized image) is marked for feature extraction until the re-extracted feature area is within the feature area of the image to be tested.
[0052] Specific implementation process: For example, this embodiment takes multiple myeloma as an example.
[0053] Initially, the image acquisition module acquires the image information of the digital pathological sections of multiple myeloma, and then the preprocessing module performs denoising on the acquired image information through median filtering to obtain a clearer image. Then the image segmentation module is used to segment the denoised image, specifically dividing the image into R single-channel image, G single-channel image, B single-channel image and binary image. Then the extraction module performs feature extraction on the R single-channel image, G single-channel image, B single-channel image and binary image respectively and marks the feature area in each image. After the feature areas in each peeled image are marked and divided, the fusion comparison module superimposes the marked R single-channel image, G single-channel image, B single-channel image and binary image according to pixel coordinates to obtain a composite image, so as to ensure that the areas of each image correspond to each other, avoid misalignment of the overlapped images, and affect the subsequent feature recognition. At the same time, the fusion comparison module also divides out the overlapping parts of the feature areas in the marked R single-channel image, G single-channel image, B single-channel image and binary image, and deepens the edge contour of the overlapping area to indicate its importance, so as to improve the recognition of the part. The recognition module classifies and recognizes the characteristic areas in the composite image using a classification recognition algorithm, wherein the classification recognition algorithm used by the recognition module is an algorithm that has been trained in advance (ie, it can recognize tumor characteristics).
[0054] Among them, the fusion comparison module also compares the feature area of the composite image including the overlapping part and the non-overlapping part with the feature area of the image to be tested, that is, when the extraction mark module extracts features of the R single-channel image, the G single-channel image, the B single-channel image, and the binary image respectively, the feature area of the image to be tested (that is, before peeling) is also extracted at the same time, and the correction module determines whether the feature area in the composite image is within the feature area in the image to be tested. If both are within the feature area in the image to be tested, it means that the features in the composite image are more precise than the features in the image to be tested. There may be areas that are difficult to distinguish and easily confused, resulting in feature recognition errors, and the feature recognition results need to be further determined. Then, the images that need to be re-featured extracted (R single-channel image, G single-channel image, B single-channel image in the composite image, which can be used as the basis for subsequent tumor identification. If they are outside the feature area in the image to be tested, they represent the special channel image and binary image in the composite image, the same below) are amplified by the magnification module according to the preset gradient. Among them, the image outside the feature area in the image to be tested may be just one of the R single-channel image, G single-channel image, B single-channel image, and binary image (that is, not all the feature areas of the images are outside the feature area in the image to be tested). Only those that meet the correction requirements need to be re-featured extracted and marked by the extraction and marking module. If the feature area of the image after the first amplification is still outside the feature area in the image to be tested, continue to amplify according to the preset gradient, for example, the first time is amplified at 40 times the speed, the second time is amplified at 100 times the speed, and so on, until the feature area in the composite image is within the feature area in the image to be tested.
[0055] Example 2
[0056] The only difference from the above embodiment is that an operation method for tumor identification based on artificial intelligence is also provided, comprising the following steps:
[0057] Step 1: Obtain the image to be tested; the image to be tested is specifically a color image JPG format, that is, a high-resolution digital image representing the entire tissue slice is obtained from a glass slide on a computer monitor through WSI technology. After the entire slide is scanned, the color image is collected.
[0058] Step 2: The image to be tested is split into R channel image, G channel image, B channel image and binary image; the image to be tested is read by the imread function, and the color image is split into three single channel images, namely, original color images, specifically, R single channel image, G single channel image and B single channel image, by the split function. At the same time, the image to be tested is binarized to obtain a binary image.
[0059] Step 3: extract and identify the features of the R channel image, the G channel image, the B channel image, the binary image and the image to be tested, and mark the feature areas respectively;
[0060] Step 4: Overlap and compare the marked R channel image, G channel image, B channel image and binary image according to the pixel coordinates in the image to be tested to obtain a fused composite image; at the same time, the feature area is divided into overlapping areas and non-overlapping areas, and the edges and contours of the overlapping areas are deepened.
[0061] Step 5: Classify and identify the composite image.
[0062] The above is only an embodiment of the present invention, and the common knowledge such as the known specific structure and / or characteristics in the scheme is not described in detail here. It should be pointed out that for those skilled in the art, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.
Claims
1. A tumor recognition system based on artificial intelligence, characterized by: Image acquisition module, image segmentation module, extraction and marking module, fusion and comparison module and recognition module; The image acquisition module is used to obtain the image to be tested; The image segmentation module is used to separate the image to be tested into a binary image and a primary color image; The extraction and marking module is used to extract and mark the features in the peeled image respectively; The fusion comparison module is used to overlap and fuse the binary image and the primary color image to obtain a composite image; The recognition module is used to classify and recognize the feature areas in the composite image through a classification recognition algorithm.
2. The artificial intelligence-based tumor recognition system according to claim 1, characterized in that: It also includes a correction module; the correction module is used to compare the feature area in the composite image with the feature area in the image to be tested, and determine whether the feature area in the composite image is within the feature area in the image to be tested. If the feature area in the composite image is within the feature area in the image to be tested, the composite image is output to the recognition module; if the feature area in the composite image is outside the feature area in the image to be tested, the features in the peeled image are re-extracted and marked until the feature area in the composite image is within the feature area in the image to be tested.
3. The artificial intelligence-based tumor recognition system according to claim 1, characterized in that: It also includes a pre-processing module; The preprocessing module is used to perform denoising on the image to be tested.
4. The artificial intelligence-based tumor recognition system according to claim 3, characterized in that: Features include texture features, shape features, and texture features.
5. The artificial intelligence-based tumor recognition system according to claim 4, characterized in that: The fusion comparison module is also used to identify the layer in the composite image that is outside the feature area of the image to be tested, and re-extract features of the layer separately; the layer is the layer where the binary image or the original color image is located.
6. The artificial intelligence-based tumor recognition system according to claim 5, characterized in that: It also includes an amplification module; the amplification module is used to receive the binary image and the original color image whose features need to be re-extracted, and amplify them according to a preset gradient speed, and send the amplified binary image and the original color image to the extraction mark module.
7. The artificial intelligence-based tumor recognition system according to claim 6, characterized in that: The fusion comparison module is also used to divide the fused feature areas into overlapping areas and non-overlapping areas, and deepen the edges and contours of the overlapping areas.
8. The artificial intelligence-based tumor recognition system according to claim 7, characterized in that: The preset gradient speeds are 40 times, 100 times, 200 times and 400 times.
9. The artificial intelligence-based tumor recognition system according to claim 8, characterized in that: The primary color image includes an R channel image, a G channel image, and a B channel image.
10. An operation method for tumor identification based on artificial intelligence, characterized in that: The following steps are involved: Step 1: Get the image to be tested; Step 2: split the image to be tested into an R channel image, a G channel image, a B channel image and a binary image; Step 3: extract and identify the features of the R channel image, the G channel image, the B channel image, the binary image and the image to be tested, and mark the feature areas respectively; Step 4: Overlap and compare the marked R channel image, G channel image, B channel image and binary image according to the pixel coordinates in the image to be tested to obtain a fused composite image; Step 5: Classify and identify the composite image.