Tumor image region extraction method and system based on neighborhood pixel feature classifier

Abstract

The application provides a tumor image region extraction method and system based on a neighborhood pixel feature classifier, comprising: collecting a medical sample polarization image of a tumor to be identified; pre-processing the medical sample polarization image to obtain a pre-processed polarization image; calculating a Mueller matrix and derived polarization feature parameters of each pixel in the pre-processed polarization image; determining a neighborhood pixel feature classifier, screening non-boundary pixels in combination with the Mueller matrix and the derived polarization feature parameters, and performing interpolation processing on the non-boundary pixels to form a multi-dimensional feature matrix; using a multi-layer perception to perform neural network identification on the multi-dimensional feature matrix, adopting four-fold cross-validation of a probability result of the neighborhood pixel feature classifier, and determining a tumor identification result to be identified based on the probability result. The application solves the problem of insensitivity of a multi-layer perception to a tumor incisal edge by using a two-dimensional feature relationship of each pixel and surrounding pixels, and realizes automatic identification of a bone tumor region and a tumor incisal edge of a fresh sample polarization image.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for extracting tumor image regions based on a neighborhood pixel feature classifier. Background Technology

[0002] In the field of medical image processing, effective identification of tumor images is of great significance, especially for some malignant tumors, such as osteosarcoma, a primary malignant tumor occurring in the skeletal system. Normally, bone mesenchymal cells differentiate into osteocytes, but due to various factors, these mesenchymal cells transform into osteosarcoma cells. Osteosarcoma typically occurs in childhood and adolescence, primarily affecting rapidly growing bone regions, commonly found in the epiphyseal regions of long bones, such as the distal femur, proximal tibia, and proximal humerus. Its incidence is higher in people aged 10 to 30, but it can also affect older adults with bone diseases or those who have undergone radiation therapy. Surgical resection is the standard clinical treatment for osteosarcoma. A significant challenge in osteosarcoma resection surgery is how to completely remove the lesion area while preserving as much of the original organ function as possible, thereby protecting the patient's organ function and improving the cure rate. During surgery, the non-cancerous tissue surrounding the excised area is called the tumor margin. If the tumor margin is misjudged, tumor cells may spread to the unexcised area, which may lead to recurrent tumors and secondary tumors. Therefore, it is very important to accurately identify the tumor area, distinguish it from the normal area, and determine the most appropriate tumor margin.

[0003] To date, surgeons primarily differentiate tumor and normal tissue areas intraoperatively through visual examination and tissue palpation, then estimate the tumor margins based on expert experience. Confirmation of complete tumor resection relies on the report of tissue sections, a time-consuming process dependent on the pathologist's expertise. With advancements in computer and artificial intelligence technologies, machine learning and deep learning have become relatively mature methods for segmenting and classifying stained pathological sections. However, these methods only identify stained pathological sections and are not well-suited for intraoperative application.

[0004] Therefore, it is of great significance to develop a rapid, accurate, and non-contact method for detecting tumor regions and tumor margins. Summary of the Invention

[0005] This invention provides a tumor image region extraction method and system based on a neighborhood pixel feature classifier, which solves the defects of low accuracy and low efficiency in the automatic identification of bone tumor regions and tumor margins in fresh samples in the prior art.

[0006] In a first aspect, the present invention provides a method for extracting tumor image regions based on a neighborhood pixel feature classifier, comprising:

[0007] Acquire polarization images of medical samples containing tumors to be identified;

[0008] The polarization image of the medical sample is preprocessed to obtain a preprocessed polarization image;

[0009] Calculate the Mueller matrix and derived polarization feature parameters for each pixel in the preprocessed polarization image;

[0010] A neighborhood pixel feature classifier is determined, and non-boundary pixels are screened by combining the Mueller matrix and the derived polarization feature parameters. The non-boundary pixels are then interpolated to form a multi-dimensional feature matrix.

[0011] The multi-dimensional feature matrix is ​​identified by a neural network using a multilayer perceptron, and the probability results of the neighborhood pixel feature classifier are verified by four-fold cross-validation. Based on the probability results, the tumor identification result is determined.

[0012] According to the present invention, a tumor image region extraction method based on a neighborhood pixel feature classifier is provided, which acquires a polarization image of a medical sample of the tumor to be identified, including:

[0013] Based on the dual-rotation quarter-wave plate method, multiple polarization images of the medical sample to be identified are acquired using a CCD.

[0014] According to the present invention, a tumor image region extraction method based on a neighborhood pixel feature classifier preprocesses the polarization image of the medical sample to obtain a preprocessed polarization image, including:

[0015] The polarization image of the medical sample is cropped to obtain the main image portion of the polarization image of the medical sample, and the background portion corresponding to the image edge of the main image portion is output as the preprocessed polarization image.

[0016] According to the present invention, a tumor image region extraction method based on a neighborhood pixel feature classifier calculates the Mueller matrix and derived polarization feature parameters of each pixel in the preprocessed polarization image, including:

[0017] The first element of the Mueller matrix is ​​determined to be the total light intensity of each pixel. Based on the first element, the remaining elements of the Mueller matrix are normalized to obtain 15 effective elements representing the polarization characteristics of the sample.

[0018] Calculate the four derived polarization feature parameters for each pixel. If an outlier occurs during the calculation, mark the current pixel, assign it a null value, and skip the current pixel. Use the 15 valid elements representing the sample polarization characteristics of all pixels and the four derived polarization feature parameters to form 19 feature parameters for each pixel, convert them into a 1×19 feature vector, and store the pixel coordinates corresponding to the outlier values ​​in an empty array.

[0019] According to the present invention, a tumor image region extraction method based on a neighborhood pixel feature classifier is provided, comprising: determining a neighborhood pixel feature classifier; combining the Mueller matrix and the derived polarization feature parameters to filter non-boundary pixels; and performing interpolation processing on the non-boundary pixels to form a multi-dimensional feature matrix, including:

[0020] Determine the size of the neighborhood window Around each cell The pixels in the large and small windows are used as neighborhood features;

[0021] Determine any pixel Whether it belongs to the boundary cell, let , ,like and If all values ​​are positive and the distance from the boundary is greater than or equal to a preset number of pixels, then the pixel is determined. For non-boundary cells, otherwise, if and If at least one negative value exists, then the cell is determined. For boundary cells;

[0022] Determine the counter Recalculate and If the calculated new ( If it is an empty array, then Any pixel to be completed All and Calculation, check Value, if Then, the value of the corresponding empty cell is estimated by interpolation. If so, no interpolation processing will be performed on the data in the window;

[0023] Around each non-boundary cell The pixel features in the large and small windows are combined, and the empty pixels appearing in the windows are added to the dimension after cubic spline interpolation. In the feature vector, no feature vector is constructed for boundary cells. After all cells are constructed, a feature vector is formed. The multidimensional feature matrix, and These represent the number of rows and columns of all pixels, respectively.

[0024] According to the present invention, a method for tumor image region extraction based on a neighborhood pixel feature classifier is provided. This method utilizes a multilayer perceptron to perform neural network recognition on the multidimensional feature matrix, employs four-fold cross-validation to verify the probability results of the neighborhood pixel feature classifier, and determines the tumor identification result based on the probability results. The method includes:

[0025] The label for a tumor cell is set to 1, and the label for a normal tissue cell is set to 0.

[0026] The training set is divided into four folds, with three folds used for training and one fold used for testing. This process is repeated four times. The false positive rate and true positive rate are calculated. After the recognition is completed, the probability result of each test image is output. The probability result is between 0 and 1.

[0027] Determine the main threshold and the floating range, and select the corresponding pixels as tumor edges from the probability map based on the main threshold and the floating range;

[0028] An image edge detection algorithm is used to extract the contour lines of the tumor edge. The extracted contour lines are then overlaid onto the original probability map in a visual manner and visualized to obtain the tumor identification result.

[0029] Secondly, the present invention also provides a tumor image region extraction system based on a neighborhood pixel feature classifier, comprising:

[0030] The acquisition module is used to acquire polarization images of medical samples containing tumors to be identified.

[0031] The preprocessing module is used to preprocess the polarization image of the medical sample to obtain a preprocessed polarization image;

[0032] The calculation module is used to calculate the Mueller matrix and derived polarization feature parameters of each pixel in the preprocessed polarization image;

[0033] The filtering module is used to determine the neighborhood pixel feature classifier, filter non-boundary pixels by combining the Mueller matrix and the derived polarization feature parameters, and perform interpolation processing on the non-boundary pixels to form a multi-dimensional feature matrix.

[0034] The identification module is used to perform neural network identification on the multidimensional feature matrix using a multilayer perceptron, and to verify the probability results of the neighborhood pixel feature classifier using four-fold cross-validation, and to determine the identification result of the tumor to be identified based on the probability results.

[0035] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the tumor image region extraction method based on a neighborhood pixel feature classifier as described above.

[0036] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the tumor image region extraction method based on a neighborhood pixel feature classifier as described above.

[0037] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the tumor image region extraction method based on a neighborhood pixel feature classifier as described above.

[0038] The tumor image region extraction method and system based on a neighborhood pixel feature classifier provided by this invention fully considers the special characteristics of bone tumors and the requirements of computing power and cost. The bone tumor region extraction accuracy of polarized images using the method of this invention is higher than that of general neural network training methods, and the computing power requirement is significantly lower than that of deep learning network training methods. It has the advantages of simple calculation, high recognition efficiency and high accuracy. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 This is one of the flowcharts illustrating the tumor image region extraction method based on a neighborhood pixel feature classifier provided by the present invention;

[0041] Figure 2 This is the second flowchart of the tumor image region extraction method based on a neighborhood pixel feature classifier provided by the present invention;

[0042] Figure 3 This is a schematic diagram of the classifier structure provided by the present invention;

[0043] Figure 4 This is a schematic diagram of the structure of the tumor image region extraction system based on a neighborhood pixel feature classifier provided by the present invention;

[0044] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0046] Figure 1 This is one of the flowcharts illustrating the tumor image region extraction method based on a neighborhood pixel feature classifier provided in this embodiment of the invention, such as... Figure 1 As shown, it includes:

[0047] Step 100: Acquire polarization images of the medical sample containing the tumor to be identified;

[0048] Step 200: Preprocess the polarization image of the medical sample to obtain a preprocessed polarization image;

[0049] Step 300: Calculate the Mueller matrix and derived polarization feature parameters for each pixel in the preprocessed polarization image;

[0050] Step 400: Determine the neighborhood pixel feature classifier, combine the Mueller matrix and the derived polarization feature parameters to filter non-boundary pixels, and perform interpolation processing on the non-boundary pixels to form a multi-dimensional feature matrix;

[0051] Step 500: Use a multilayer perceptron to perform neural network recognition on the multidimensional feature matrix, use four-fold cross-validation to verify the probability results of the neighborhood pixel feature classifier, and determine the tumor identification result based on the probability results.

[0052] This invention employs a neighborhood pixel feature classifier based on polarized light to extract tumor image regions. It is understood that under polarized light illumination, objects with different structures scatter light differently depending on the polarization. Therefore, polarized images can provide important information about the surface properties, material, and structure of objects. In modern medicine, polarization imaging is a non-destructive, label-free, highly sensitive, and non-invasive optical detection technique. The Mueller matrix and its derived parameters are a relatively complete method for characterizing the polarization information of biological tissues, often referred to as the "fingerprint" of biological tissue optical properties. Traditional Mueller matrix measurement methods primarily analyze the characteristics of the Mueller matrix and related derived parameters, seeking differences between tumors and other tissues in the Mueller matrix elements and derived parameters, and then distinguishing tumor regions from normal tissue regions by setting thresholds. However, relying solely on existing Mueller matrix and derived parameter analysis methods cannot effectively distinguish between osteosarcoma and tumor margins. This is because fresh medical samples have a more complex structure than pathological sections, resulting in more complex absorption and scattering of polarized light, which cannot be explained by a single scattering principle. Machine learning methods can effectively learn all parameter features, iteratively refining them as the training set size increases. For example, multilayer perceptrons (MLPs) can be used to extract and classify features from the Mueller matrix and derived parameters of polarization images. The principle of an MLP is to learn features and classify each pixel independently. However, learning from a single pixel using an MLP cannot capture the regional clustering and boundary diffusion characteristics of osteosarcoma cancer cells. Therefore, the method employed in this invention, by emphasizing the two-dimensional feature relationships between each pixel and its surrounding pixels, can overcome the insensitivity of MLPs to tumor margins, thus enabling automatic identification of bone tumor regions and tumor margins in fresh polarization images.

[0053] The method provided by this invention can be implemented using computer software technology. The embodiment uses a polarization image of a fresh mouse bone tumor sample as an example to illustrate the process of this invention.

[0054] In one embodiment, step 100 includes:

[0055] Polarization images of fresh medical samples were acquired using an optical measurement device. Based on the double-rotated quarter-wave plate method, discrete images were acquired using a CCD, resulting in 32 polarization images for a single fresh sample.

[0056] In one embodiment, step 200 includes:

[0057] Data preprocessing is performed to eliminate errors caused by instrument movement during shooting, light intensity errors due to changes in ambient light intensity, and image cropping where fresh samples occupy the majority of the image. During image cropping, it should be ensured that the sample constitutes the majority of each image, and that the edges of the image are background, not samples exhibiting polarization characteristics.

[0058] In one embodiment, step 300 includes:

[0059] Calculate the Mueller matrix and four derived polarization feature parameters for each pixel. For each pixel... Calculate its corresponding Mueller matrix. The Mueller matrix is ​​a... The matrix contains 16 elements. The first element of the Mueller matrix. This reflects the total light intensity of that pixel. Next, we will examine the Mueller matrix. Except Normalize all other elements except those listed above:

[0060]

[0061] for ,and This yields 15 effective elements from the Mueller matrix that reflect the polarization characteristics of the sample.

[0062] The derived polarization parameters are then calculated. During this calculation, outliers or unstable results may be encountered. If certain pixels... If a polarization parameter exhibits an outlier value, the data for that pixel is assigned a null value, the calculation of its derived parameters is skipped, and its coordinates are stored in an array, denoted as . .

[0063] In one embodiment, step 400 includes:

[0064] During cancer cell growth, cancerous tissue spreads outward from the cancer nest, resulting in a dense distribution of cancer cells near the cancer nest area with relatively blurred boundaries. Therefore, a classifier based on neighborhood pixel features is set, and the neighborhood window size is determined. In addition to the edges of the image, the area surrounding each cell is also considered. Cells within the large and small windows serve as neighborhood features. In the analysis of each cell, the first step is to determine whether the cell belongs to the boundary cell. For example... Figure 2 As shown, the judgment method is as follows:

[0065] Let the window size be And let the pixel being judged be Then for this pixel, there are several possibilities: (1) Pixel (1) Located at the upper or lower boundary of the image; (2) Pixel Located at the left or right boundary of the image; (3) pixel Not located at the boundary of the image, but less than 5 pixels away from the boundary; (4) pixels It is not located on the image boundary, and its distance from the boundary is greater than or equal to 5 pixels. Therefore, let... and set .in , For pixels The coordinates. If and If all values ​​are positive, then the pixel can be considered as a pixel. Not located on the image boundary, and at least 5 pixels away from the boundary; conversely, if and If at least one negative value exists, then the pixel is considered to be... It is within the range of the boundary cell.

[0066] After confirming that a cell is a non-boundary cell, the following determination is made for that cell:

[0067] Let the pixel being judged be and set a counter Recalculate and Each set of calculations Then check ( Does it belong to the empty pixel set obtained in step 3? .like There exists (in) Then make In pixels All and After all calculations are completed, check. Size: If If the number of empty pixels is less than half the total number of pixels in the window excluding the current pixel, then the data within the window is considered sufficiently continuous, and the values ​​of these empty pixels can be estimated using interpolation methods; if If a certain number of data points are found in the window, the data in that window will be considered to be affected by factors such as specular reflection, making it impossible to correctly express the polarization characteristics of the material through the calculation of the Mueller matrix and derived parameters. Therefore, the data in that window will not be interpolated.

[0068] For all non-boundary pixels that meet the above criteria, the area around each pixel will be... The pixel features in the large and small windows are combined, and based on the 15 Mueller matrix elements and 4 derived polarization feature parameters calculated in step 3, a latitude-valued array is formed. The feature vector. In constructing the feature vector, empty pixels appearing in the window are interpolated using cubic splines and then added to the feature vector at latitude . In the feature vector, this interpolation operation only changes the value of the empty cell as a neighborhood feature, not its value in the original matrix. For boundary cells, no feature vector is constructed. The above steps are performed on all cells to form a... The multidimensional feature matrix, such as Figure 3 As shown.

[0069] In one embodiment, step 500 includes:

[0070] After interpolation, a multilayer perceptron (MLP) is used for neural network recognition. A four-fold cross-validation method is employed to evaluate the performance of the neighborhood pixel-based classifier. Pixels containing tumors are labeled as 1 (positive), and pixels containing normal tissue are labeled as 0 (negative). The training set is divided into four folds, with three folds used for training and one fold for testing, repeated four times to calculate the false positive and true positive rates. After recognition, the probability segmentation result for each test image is output. The probability value for the segmentation result is between 0 and 1. A primary threshold of 0.5 and a floating range of ±0.1 are set to define the possible tumor region. Pixels with probability values ​​within the 0.5 ± 0.1 range are selected from the probability map and considered as tumor margins. Subsequently, an image edge detection algorithm is used to accurately depict the contour line of the tumor margin. The extracted contour line is then visually overlaid onto the original probability map and visualized as the recognition result.

[0071] The tumor image region extraction system based on a neighborhood pixel feature classifier provided by the present invention is described below. The tumor image region extraction system based on a neighborhood pixel feature classifier described below can be referred to in correspondence with the tumor image region extraction method based on a neighborhood pixel feature classifier described above.

[0072] Figure 4 This is a schematic diagram of the structure of the tumor image region extraction system based on a neighborhood pixel feature classifier provided in an embodiment of the present invention, as shown below. Figure 4 As shown, it includes: a data acquisition module 41, a preprocessing module 42, a calculation module 43, a filtering module 44, and a recognition module 45, wherein:

[0073] The acquisition module 41 is used to acquire polarization images of medical samples of the tumor to be identified; the preprocessing module 42 is used to preprocess the polarization images of the medical samples to obtain preprocessed polarization images; the calculation module 43 is used to calculate the Mueller matrix and derived polarization feature parameters of each pixel in the preprocessed polarization image; the filtering module 44 is used to determine the neighborhood pixel feature classifier, filter non-boundary pixels by combining the Mueller matrix and the derived polarization feature parameters, and perform interpolation processing on the non-boundary pixels to form a multidimensional feature matrix; the recognition module 45 is used to perform neural network recognition on the multidimensional feature matrix using a multilayer perceptron, and use four-fold cross-validation to verify the probability results of the neighborhood pixel feature classifier, and determine the tumor identification result based on the probability results.

[0074] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a tumor image region extraction method based on a neighborhood pixel feature classifier. This method includes: acquiring a polarization image of a medical sample of the tumor to be identified; preprocessing the polarization image of the medical sample to obtain a preprocessed polarization image; calculating the Mueller matrix and derived polarization feature parameters of each pixel in the preprocessed polarization image; determining a neighborhood pixel feature classifier, combining the Mueller matrix and the derived polarization feature parameters to filter non-boundary pixels, interpolating the non-boundary pixels to form a multidimensional feature matrix; using a multilayer perceptron to perform neural network recognition on the multidimensional feature matrix, using four-fold cross-validation to verify the probability results of the neighborhood pixel feature classifier, and determining the tumor identification result based on the probability results.

[0075] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0076] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the tumor image region extraction method based on a neighborhood pixel feature classifier provided by the above methods. The method includes: acquiring a polarization image of a medical sample of a tumor to be identified; preprocessing the polarization image of the medical sample to obtain a preprocessed polarization image; calculating the Mueller matrix and derived polarization feature parameters of each pixel in the preprocessed polarization image; determining a neighborhood pixel feature classifier, filtering non-boundary pixels by combining the Mueller matrix and the derived polarization feature parameters, interpolating the non-boundary pixels to form a multidimensional feature matrix; using a multilayer perceptron to perform neural network recognition on the multidimensional feature matrix, using four-fold cross-validation to verify the probability results of the neighborhood pixel feature classifier, and determining the tumor identification result based on the probability results.

[0077] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a tumor image region extraction method based on a neighborhood pixel feature classifier provided by the above methods. This method includes: acquiring a polarization image of a medical sample of a tumor to be identified; preprocessing the polarization image of the medical sample to obtain a preprocessed polarization image; calculating the Mueller matrix and derived polarization feature parameters for each pixel in the preprocessed polarization image; determining a neighborhood pixel feature classifier, filtering non-boundary pixels by combining the Mueller matrix and the derived polarization feature parameters, interpolating the non-boundary pixels to form a multidimensional feature matrix; performing neural network recognition on the multidimensional feature matrix using a multilayer perceptron, verifying the probability results of the neighborhood pixel feature classifier using four-fold cross-validation, and determining the tumor identification result based on the probability results.

[0078] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0079] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for extracting tumor image regions based on a neighborhood pixel feature classifier, characterized in that, include: Acquire polarization images of medical samples containing tumors to be identified; The polarization image of the medical sample is preprocessed to obtain a preprocessed polarization image; Calculate the Mueller matrix and derived polarization feature parameters for each pixel in the preprocessed polarization image; A neighborhood pixel feature classifier is determined, and non-boundary pixels are screened by combining the Mueller matrix and the derived polarization feature parameters. The non-boundary pixels are then interpolated to form a multi-dimensional feature matrix. The multi-dimensional feature matrix is ​​identified by a neural network using a multi-layer perceptron, and the probability results of the neighborhood pixel feature classifier are verified by four-fold cross-validation. The tumor identification result is determined based on the probability results. A neighborhood pixel feature classifier is determined, and non-boundary pixels are filtered by combining the Mueller matrix and the derived polarization feature parameters. Interpolation processing is then performed on the non-boundary pixels to form a multi-dimensional feature matrix, including: Determine the size of the neighborhood window Around each cell The pixels in the large and small windows are used as neighborhood features; Determine any pixel Whether it belongs to the boundary cell, let , ,like and If all values ​​are positive and the distance from the boundary is greater than or equal to a preset number of pixels, then the pixel is determined. For non-boundary cells, otherwise, if and If at least one negative value exists, then the cell is determined. For boundary cells; Determine the counter Recalculate and If the calculated new ( If it is an empty array, then Any pixel to be completed All and Calculation, check Value, if Then, the value of the corresponding empty cell is estimated by interpolation. If so, no interpolation processing will be performed on the data in the window; Around each non-boundary cell The pixel features in the large and small windows are combined, and the empty pixels appearing in the windows are added to the dimension after cubic spline interpolation. In the feature vector, no feature vector is constructed for boundary cells. After all cells are constructed, a feature vector is formed. The multidimensional feature matrix, and These represent the number of rows and columns of all pixels, respectively.

2. The tumor image region extraction method based on a neighborhood pixel feature classifier according to claim 1, characterized in that, Acquire polarization images of the medical sample containing the tumor to be identified, including: Based on the dual-rotation quarter-wave plate method, multiple polarization images of the medical sample to be identified are acquired using a CCD.

3. The tumor image region extraction method based on a neighborhood pixel feature classifier according to claim 1, characterized in that, The polarization image of the medical sample is preprocessed to obtain a preprocessed polarization image, including: The polarization image of the medical sample is cropped to obtain the main image portion of the polarization image of the medical sample, and the background portion corresponding to the image edge of the main image portion is output as the preprocessed polarization image.

4. The tumor image region extraction method based on a neighborhood pixel feature classifier according to claim 1, characterized in that, Calculate the Mueller matrix and derived polarization feature parameters for each pixel in the preprocessed polarization image, including: The first element of the Mueller matrix is ​​determined to be the total light intensity of each pixel. Based on the first element, the remaining elements of the Mueller matrix are normalized to obtain 15 effective elements representing the polarization characteristics of the sample. Calculate the four derived polarization feature parameters for each pixel. If an outlier occurs during the calculation, mark the current pixel, assign it a null value, and skip the current pixel. Use the 15 valid elements representing the sample polarization characteristics of all pixels and the four derived polarization feature parameters to form 19 feature parameters for each pixel, convert them into a 1×19 feature vector, and store the pixel coordinates corresponding to the outlier values ​​in an empty array.

5. The tumor image region extraction method based on a neighborhood pixel feature classifier according to claim 1, characterized in that, The multi-dimensional feature matrix is ​​subjected to neural network recognition using a multilayer perceptron. The probability results of the neighborhood pixel feature classifier are validated using four-fold cross-validation. Based on the probability results, the tumor identification result is determined, including: The label for a tumor cell is set to 1, and the label for a normal tissue cell is set to 0. The training set is divided into four folds, with three folds used for training and one fold used for testing. This process is repeated four times. The false positive rate and true positive rate are calculated. After the recognition is completed, the probability result of each test image is output. The probability result is between 0 and 1. Determine the main threshold and the floating range, and select the corresponding pixels as tumor edges from the probability map based on the main threshold and the floating range; An image edge detection algorithm is used to extract the contour lines of the tumor edge. The extracted contour lines are then overlaid onto the original probability map in a visual manner and visualized to obtain the tumor identification result.

6. A tumor image region extraction system based on a neighborhood pixel feature classifier, based on the tumor image region extraction method based on a neighborhood pixel feature classifier according to any one of claims 1 to 5, characterized in that, include: The acquisition module is used to acquire polarization images of medical samples containing tumors to be identified. The preprocessing module is used to preprocess the polarization image of the medical sample to obtain a preprocessed polarization image; The calculation module is used to calculate the Mueller matrix and derived polarization feature parameters of each pixel in the preprocessed polarization image; The filtering module is used to determine the neighborhood pixel feature classifier, filter non-boundary pixels by combining the Mueller matrix and the derived polarization feature parameters, and perform interpolation processing on the non-boundary pixels to form a multi-dimensional feature matrix. The identification module is used to perform neural network identification on the multidimensional feature matrix using a multilayer perceptron, and to verify the probability results of the neighborhood pixel feature classifier using four-fold cross-validation, and to determine the identification result of the tumor to be identified based on the probability results.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the tumor image region extraction method based on a neighborhood pixel feature classifier as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the tumor image region extraction method based on a neighborhood pixel feature classifier as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the tumor image region extraction method based on a neighborhood pixel feature classifier as described in any one of claims 1 to 5.

Images