Breast tissue pathological image classification method and system

Abstract

The invention discloses a mammary tissue pathological image classification method and system, and relates to the technical field of mammary tissue pathological image classification. An image acquisition module is used for acquiring a mammary tissue pathological image, and an image processing module is used for performing feature extraction processing on the acquired mammary tissue pathological image; the feature extraction comprises gray value extraction, texture feature extraction and abnormal cell number extraction of pathological features, an image feature extraction index TQ, classification confidence KX and a cyclic influence coefficient XH are sequentially calculated and output based on feature extraction by using an image data calculation module, and iterative processing is carried out based on the cyclic influence coefficient XH by using an iterative analysis module, so that the classification confidence KX and the cyclic influence coefficient XH are calculated and output; according to the method, adaptive adjustment of image features is realized, the accuracy and reliability of classification are improved, the generalization ability of the model is enhanced, and a new powerful tool is provided for early diagnosis and treatment of breast cancer.

Description

Technical Field

[0001] The present invention relates to the technical field of breast tissue pathology image classification, and in particular to a breast tissue pathology image classification method and system. Background Art

[0002] Breast cancer is one of the leading causes of death among women. Its early and accurate diagnosis is crucial for treatment and prognosis. Histopathological images are an important basis for the diagnosis of breast cancer. Pathologists make diagnoses by observing various visual features in pathological sections, including the morphological characteristics of the cell nucleus and the microscopic and macroscopic structures of the cell nucleus. In recent years, with the rapid development of computer technology, image classification technology based on machine learning and deep learning has provided new solutions for the classification of breast tissue pathology images.

[0003] At present, existing technologies often rely on fixed feature extraction methods, including SIFT and SURF. However, these methods will find it difficult to accurately extract key features when processing complex and changeable breast tissue pathology images, resulting in poor classification results. Moreover, due to inaccurate extraction of image features, the classification confidence of existing technologies is often low, making it difficult to ensure the reliability of classification results. In addition, classification models in existing technologies often lack adaptive adjustment capabilities. When faced with new and unknown breast tissue pathology images, the generalization ability of the models is poor and it is difficult to accurately classify them. Moreover, classification methods in existing technologies are often one-off and lack a cyclic influence mechanism. It is difficult to adaptively adjust image features according to classification results, thereby limiting further improvement in classification accuracy. Summary of the invention

[0004] The purpose of the present invention is to provide a breast tissue pathology image classification method to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution, including a method for adaptively iteratively adjusting image features according to classification results;

[0006] The specific implementation steps are as follows:

[0007] Step 1: using an image acquisition module to acquire images of breast tissue pathology;

[0008] Step 2: Using the image processing module, the collected breast tissue pathology image is subjected to feature extraction processing, wherein the feature extraction includes gray value extraction, texture feature extraction, and abnormal cell number extraction of pathological features;

[0009] Step 3: Based on feature extraction and using the image data calculation module, the output image feature extraction index TQ, classification confidence KX, and cyclic influence coefficient XH are calculated in sequence;

[0010] Step 4: Based on the cyclic influence coefficient XH, iterative processing is performed using an iterative analysis module;

[0011] The image data calculation module includes a unit for comprehensively reflecting the overall characteristics of the image, a unit for reflecting the reliability of the image being correctly classified, and a unit for cyclically and optimizing the image feature extraction.

[0012] Optionally, the equipment used by the image acquisition module includes a microscope and a scanner;

[0013] The equipment used in the image processing module includes medical image processing software and professional pathology analysis tools;

[0014] The equipment used by the image data calculation module includes a high-performance computer;

[0015] The equipment used by the iterative analysis module includes iterative control and analysis tools.

[0016] Optionally, the calculation formula for comprehensively reflecting the overall feature unit of the image is as follows:

[0017]

[0018] in:

[0019] TQ is the image feature extraction index;

[0020] N is the total number of pixels in the image;

[0021] H i is the gray value of the i-th pixel;

[0022] W i is the texture feature value of the i-th pixel;

[0023] BL i is the number of abnormal cells in the i-th pixel.

[0024] Optionally, the i-th pixel texture feature value W i The calculation formula is as follows:

[0025]

[0026] h is the number of gray levels;

[0027] a and b are grayscale values. a and b represent the grayscale values ​​of two pixels at a certain distance in the image, and the two pixels are adjacent;

[0028] HG(a,b) is the value of the element (a,b) in the gray-level co-occurrence matrix, which represents the joint probability that two pixels at a certain distance in the image have gray-level values ​​a and b respectively;

[0029] Among them, if the image is an 8-bit grayscale image, then the number of gray levels h is 256 (from 0 to 255). When calculating the grayscale co-occurrence matrix, it is necessary to introduce the grayscale values ​​of two pixels in the image that are a certain distance apart, namely a and b. Since the grayscale value is an integer from 0 to h-1, h-1 represents the maximum grayscale value.

[0030] Optionally, the calculation formula for the reliability unit reflecting the correct classification of the image is as follows:

[0031] KX=TQ×BC H -[(BL max -BL min ) / (TQ+SQRT(BC W ))];

[0032] in:

[0033] KX is the classification confidence;

[0034] BC H is the standard deviation of the image gray value;

[0035] BL max is the maximum number of abnormal cells;

[0036] BL min is the minimum number of abnormal cells;

[0037] BC W is the standard deviation of the image texture eigenvalues.

[0038] Optionally, the calculation formula of the loop and optimization image feature extraction unit is as follows:

[0039] XH=[(KX×(TQ+SQRT(BL avg ))) / (1+(BC BL / TQ) 2 )]-((KX-TQ) / N);

[0040] in:

[0041] XH is the circulation influence coefficient;

[0042] BL avg is the average number of abnormal cells;

[0043] BC BL is the standard deviation of the number of abnormal cells in the image.

[0044] Optionally, the adjustment calculation formula based on the circulation influence coefficient XH is as follows:

[0045] TQ new=TQ+XH;

[0046] in:

[0047] TQ new is the new round of image feature extraction index, and the new round of image feature extraction index TQ new The unit reflecting the reliability of correct classification of the image and the unit for extracting the loop and optimized image features are introduced in sequence to form an iterative loop. The specific method for adjusting the iterative loop and stopping the iteration is as follows:

[0048] If the cyclic influence coefficient XH is 0, it means that the classification result has no feedback effect on the image feature extraction index TQ. When combined with the new round of image feature extraction index TQ new Carry out a new round of iteration;

[0049] If the cyclic influence coefficient XH is a positive number, it reflects that the classification result has a positive feedback effect on the image feature extraction index TQ, and when the result of the cyclic influence coefficient XH continues to approach 1, the iteration is stopped. In addition, if the cyclic influence coefficient XH is a positive number and remains a positive value during the iteration process, the iteration is also stopped;

[0050] If the cyclic influence coefficient XH is negative, it means that the classification result has a negative feedback effect on the image feature extraction index TQ, so the classifier parameters should be checked and adjusted.

[0051] Optionally, the image gray value standard deviation BC H The calculation formula is as follows:

[0052]

[0053] H avg =(H1+H2+H3+......+H N ) / N;

[0054] H avg is the average gray value of pixels;

[0055] H1 is the grayscale value of the first pixel, H2 is the grayscale value of the second pixel, H3 is the grayscale value of the third pixel, N is the gray value of the Nth pixel;

[0056] The standard deviation BC of the image texture feature value W The calculation formula is as follows:

[0057]

[0058] W avg =(W1+W2+W3+......+W N ) / N;

[0059] W avg is the average eigenvalue of pixel texture;

[0060] W1 is the first pixel texture eigenvalue, W2 is the second pixel texture eigenvalue, W3 is the third pixel texture eigenvalue, W N is the Nth pixel texture feature value;

[0061] The standard deviation BC of the number of abnormal cells in the image BL The calculation formula is as follows:

[0062]

[0063] BL avg =(BL1+BL2+BL3+......+BL N ) / N;

[0064] BL1 is the number of abnormal cells in the first pixel, BL2 is the number of abnormal cells in the second pixel, BL3 is the number of abnormal cells in the third pixel, and BL N is the number of abnormal cells in the Nth pixel.

[0065] The present invention also provides a breast tissue pathology image classification system, including an image acquisition module, an image processing module, an image data calculation module and an iterative analysis module.

[0066] Image acquisition module: responsible for collecting breast tissue pathology images;

[0067] Image processing module: responsible for feature extraction of abnormal cell numbers including grayscale value, texture features and pathological features:

[0068] Image data calculation module: responsible for calculating the output image feature extraction index TQ, classification confidence KX, and cyclic influence coefficient XH in sequence;

[0069] Iterative analysis module: responsible for iteration based on the cycle influence coefficient XH.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] 1. The present invention comprehensively reflects the overall feature units of the image and comprehensively introduces the grayscale value, texture features and the number of abnormal cells of the pathological features of the image, so as to more accurately extract the key features of the breast tissue pathology image.

[0072] 2. The present invention reflects the reliability unit of the correct classification of the image and calculates the classification confidence KX based on the image feature extraction index TQ, the standard deviation of the gray value and the specific pathological feature value, so as to more reliably evaluate the accuracy of the classification result.

[0073] 3. The present invention realizes adaptive adjustment of image features by looping and optimizing image feature extraction units and loop influence mechanisms, so that the classification model can have stronger generalization ability when facing new and unknown breast tissue pathology images. In addition, a loop influence mechanism is introduced in the classification of breast tissue pathology images, which realizes continuous optimization of image features by continuously iteratively updating the image feature extraction index TQ, thereby further improving the accuracy of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is a flow chart of the method and system for classifying breast tissue pathology images;

[0075] Figure 2 Schematic diagram of the module structure of the breast tissue pathology image classification method and system;

[0076] Figure 3 It is a structural schematic diagram of the image data calculation and processing module of the present invention;

[0077] Figure 4 A schematic diagram of the iterative condition settings of the present invention. DETAILED DESCRIPTION

[0078] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] The present breast tissue pathology image classification method and system are different from the existing breast tissue pathology image classification methods and systems. The existing breast tissue pathology image classification methods and systems have the problems of inaccurate image feature extraction, low classification confidence, poor model generalization ability and inability to achieve circular feedback. The present algorithm unit realizes adaptive adjustment of image features, improves classification accuracy and reliability, enhances the generalization ability of the model, and provides a new and powerful tool for early diagnosis and treatment of breast cancer.

[0080] For example, see Figures 1 to 4 ,This implementation provides a breast tissue pathology image classification method, including a method for adaptively iteratively adjusting image features according to the classification results;

[0081] The specific implementation steps are as follows:

[0082] Step 1: using an image acquisition module to acquire images of breast tissue pathology;

[0083] Step 2: Using the image processing module, the collected breast tissue pathology image is subjected to feature extraction processing, wherein the feature extraction includes gray value extraction, texture feature extraction, and abnormal cell number extraction of pathological features;

[0084] Step 3: Based on feature extraction and using the image data calculation module, the output image feature extraction index TQ, classification confidence KX, and cyclic influence coefficient XH are calculated in sequence;

[0085] Step 4: Based on the cyclic influence coefficient XH, iterative processing is performed using an iterative analysis module;

[0086] The image data calculation module includes a unit for comprehensively reflecting the overall characteristics of the image, a unit for reflecting the reliability of the image being correctly classified, and a unit for cyclically and optimizing the image feature extraction;

[0087] The equipment used in the image acquisition module includes microscopes and scanners;

[0088] The equipment used in the image processing module includes medical image processing software and professional pathology analysis tools;

[0089] The equipment used in the image data calculation module includes high-performance computers;

[0090] The equipment used in the iterative analysis module includes iterative control and analysis tools.

[0091] See also Figures 1 to 4 ,This implementation also provides a breast tissue pathology image classification system, including an image acquisition module, an image processing module, an image data calculation module and an iterative analysis module.

[0092] In this embodiment, the system improves the accuracy and efficiency of classification through the mutual cooperation of three algorithm units and the combination of three operation results of TQ, KX and XH. Specifically, TQ is the image feature extraction index, which can more accurately identify the key features in breast tissue pathology images, including the morphology, size and distribution of cell nuclei, and these features are of great significance for distinguishing different types of breast tissue pathology images. KX is the classification confidence, which can evaluate the accuracy of the classification results and adjust and optimize the classifier parameters according to the evaluation results, thereby improving the accuracy and efficiency of classification. At the same time, KX also serves as a basis for auxiliary decision-making to help the system in breast tissue pathology. Make more informed decisions during image analysis. XH is the cyclic influence coefficient, which continuously iterates and updates TQ through a cyclic influence mechanism, so that the classification system can continuously optimize its own performance according to the classification results, thereby forming a virtuous circle. The calculation results of XH can also affect the calculation of TQ and KX, so that the three algorithms of this system play an important role in the breast tissue pathology image classification method and system, and the three algorithms are interconnected through a cyclic influence mechanism, which jointly improves the accuracy and efficiency of classification. This cyclic influence mechanism not only optimizes the image feature extraction process, but also improves the accuracy and robustness of classification, providing strong support for the early diagnosis and treatment of breast cancer.

[0093] See also Figures 1 to 4 , the calculation formula that fully reflects the overall feature unit of the image is as follows:

[0094]

[0095] in:

[0096] TQ is the image feature extraction index;

[0097] N is the total number of pixels in the image;

[0098] H i is the gray value of the i-th pixel;

[0099] W i is the texture feature value of the i-th pixel;

[0100] BL i is the number of abnormal cells in the i-th pixel;

[0101] The texture feature value W of the i-th pixel i The calculation formula is as follows:

[0102]

[0103] h is the number of gray levels;

[0104] a and b are grayscale values. a and b represent the grayscale values ​​of two pixels at a certain distance in the image, and the two pixels are adjacent;

[0105] HG(a,b) is the value of the element (a,b) in the gray-level co-occurrence matrix, which represents the joint probability that two pixels at a certain distance in the image have gray-level values ​​a and b respectively;

[0106] Among them, if the image is an 8-bit grayscale image, then the number of gray levels h is 256 (from 0 to 255). When calculating the grayscale co-occurrence matrix, it is necessary to introduce the grayscale values ​​of two pixels in the image that are a certain distance apart, namely a and b. Since the grayscale value is an integer from 0 to h-1, h-1 represents the maximum grayscale value.

[0107] In this embodiment: First, in this algorithm unit The calculation part calculates the square root of the sum of the squares of the grayscale values ​​and texture feature values ​​of all pixels in the image. It aims to comprehensively evaluate the grayscale information and texture features of the image, so as to obtain a quantitative index that can reflect the overall characteristics of the image. As the numerator in the calculation formula that comprehensively reflects the overall characteristic unit of the image, it provides basic quantitative information of the image characteristics and is used in conjunction with the subsequent abnormal cell number BL of the i-th pixel. i After the subtraction operation, the final value of the image feature extraction index TQ is jointly determined;

[0108] The calculation part calculates the sum of the pathological feature values ​​of all pixels in the image. It aims to evaluate the overall number of pathological features in the image and provide pathological information for the calculation of the image feature extraction index TQ. As the denominator in the calculation formula that fully reflects the overall feature unit of the image, it is subtracted from the comprehensive quantitative index of the grayscale value and texture feature value to obtain the image feature extraction index TQ. The high or low image feature extraction index TQ reflects the comprehensive performance of the image in terms of grayscale, texture and pathological features;

[0109] As one of the basic attributes of the image, the grayscale value of this algorithm unit reflects the brightness information of the image. In the calculation of the image feature extraction index TQ, the grayscale value is used to measure the brightness of different areas in the image, so as to help identify areas with specific pathological characteristics. Texture is a local pattern and structure that recurs in the image, which reflects the microstructural information of the image. In the image feature extraction index TQ, texture features are used to capture subtle changes in the image, including the morphology, size and distribution of cell nuclei. These changes are of great significance for distinguishing different types of breast tissue pathology images. In addition to grayscale and texture features, the image feature extraction index TQ also considers other quantities related to pathological features. The introduction of these attributes further enriches the information content of the image feature extraction index TQ, enabling it to more comprehensively reflect the pathological information of the image.

[0110] The beneficial effect of the image feature extraction index TQ calculated and output by this algorithm is that it can comprehensively consider multiple image features, avoiding the limitations of a single feature extraction method, thereby improving the comprehensiveness and accuracy of feature extraction. At the same time, the calculation process of the image feature extraction index TQ is relatively simple and can be optimized using existing image processing techniques and algorithms, thereby improving computational efficiency.

[0111] See also Figures 1 to 4 , the calculation formula reflecting the reliability unit of the image being correctly classified is as follows:

[0112] KX=TQ×BC H -[(BL max -BL min ) / (TQ+SQRT(BC W ))];

[0113] in:

[0114] KX is the classification confidence;

[0115] BC H is the standard deviation of the image gray value;

[0116] BL max is the maximum number of abnormal cells;

[0117] BL min is the minimum number of abnormal cells;

[0118] BC W is the standard deviation of the image texture eigenvalues.

[0119] In this embodiment, first, "TQ×BC H The calculation part calculates the image feature extraction index TQ and the image gray value standard deviation BC HThe product of , which aims to evaluate the relationship between the image feature extraction index TQ and the image gray value volatility, thereby reflecting the sensitivity of the classification result to the image gray feature. As part of the addition operation in the calculation formula of the reliability unit reflecting the correct classification of the image, it and the subsequent subtraction operation jointly determine the final value of the classification confidence KX, image feature extraction index TQ × image gray value standard deviation BC H The larger the value of, the more sensitive the classification result is to the grayscale features of the image;

[0120] “[(BL max -BL min ) / (TQ+SQRT(BC W The calculation part calculates the difference between the maximum and minimum differences of the pathological feature values ​​divided by the image feature extraction index TQ and the standard deviation BC of the image texture feature value W The square root of , which is intended to evaluate the relationship between the image feature extraction index TQ and the image texture feature volatility and the pathological feature value range, thereby reflecting the sensitivity of the classification results to the image texture and pathological features. As the subtrahend part of the subtraction operation in the classification confidence KX formula, it is related to "TQ×BC H "The values ​​of the calculation part are subtracted to obtain the classification confidence KX value. The classification confidence KX value reflects the comprehensive sensitivity of the classification result to the image features including grayscale, texture and pathological features;

[0121] This algorithm unit is the result of image feature extraction. Its image feature extraction index TQ reflects the extraction of key features in the image. In the calculation of classification confidence KX, the image feature extraction index TQ is used to measure the similarity and distance between the image and a specific category.

[0122] Image gray value standard deviation BC H It reflects the fluctuation of grayscale values ​​in the image. In the classification confidence KX, the standard deviation of grayscale values ​​is used to measure the uniformity and heterogeneity of the image, which is of great significance for distinguishing different types of breast tissue pathology images.

[0123] The beneficial effect of the classification confidence KX calculated by the reliability unit reflecting the correct classification of the image is that it can reflect the reliability of the classification result, thereby helping to judge the accuracy of the classification result and reducing the risk of misdiagnosis and missed diagnosis. At the same time, the classification confidence KX can also be used as a basis for classifier performance evaluation to guide the parameter adjustment and optimization of the classifier.

[0124] See also Figures 1 to 4 , the calculation formula for looping and optimizing the image feature extraction unit is as follows:

[0125] XH=[(KX×(TQ+SQRT(BL avg))) / (1+(BC BL / TQ) 2 )]-((KX-TQ) / N);

[0126] in:

[0127] XH is the circulation influence coefficient;

[0128] BL avg is the average number of abnormal cells;

[0129] BC BL is the standard deviation of the number of abnormal cells in the image.

[0130] In this embodiment, the algorithm unit first "(BC BL / TQ) 2 The calculation part calculates the image feature extraction index TQ and the standard deviation of the number of abnormal cells in the image BC BL The square of the product of the squares of is intended to evaluate the relationship between the image feature extraction index TQ and the volatility of the pathological feature value, and to amplify the impact of this relationship through the square operation. As part of the addition operation in the calculation formula of the cyclic and optimized image feature extraction unit, it is multiplied with the classification confidence KX value to jointly determine a part of the final value of the cyclic influence coefficient XH. Figure "(BC BL / TQ) 2 "The larger the value of the calculation part, the stronger the relationship between the image feature extraction index TQ and the volatility of the pathological characteristic value, and the greater the impact on the circulation influence coefficient XH value;

[0131] “(KX×(TQ+SQRT(BL avg The calculation part calculates the classification confidence KX value, image feature extraction index TQ and the average number of abnormal cells BL avg It is designed to comprehensively evaluate the relationship between the classification confidence KX and the image features, thereby reflecting the feedback effect of the classification results on the image features. As another part of the addition operation in the calculation formula of the loop and optimization image feature extraction unit, it is related to "(BC BL / TQ) 2 "The values ​​of the calculation part are divided and then the subsequent subtraction operation is performed to jointly determine the final value of the circulation influence coefficient XH," (KX × (TQ + SQRT (BL avg The larger the value of the calculation part of )))”, the stronger the feedback effect of the classification result on the image features;

[0132] The calculation part of "((KX-TQ) / N)" calculates the difference between the classification confidence value KX and the image feature extraction index TQ divided by the total number of pixels N in the image, and is used to adjust the influence of the classification confidence value KX on the final cyclic influence coefficient XH value. As the subtrahend part of the subtraction operation in the calculation formula that fully reflects the overall feature unit of the image, it is subtracted from the previous addition operation result to obtain the cyclic influence coefficient XH value;

[0133] This algorithm unit is used as a reliability indicator of the classification result. Its classification confidence KX plays a key role in the calculation of the cyclic influence coefficient XH. It reflects the classification effect of the classifier on the current image, thereby guiding the cyclic influence coefficient XH to adaptively adjust the image features. In addition to the classification confidence KX, the cyclic influence coefficient XH can also consider other image feature values, including the mean of the grayscale value and the statistics of the texture features. The introduction of these feature values ​​further enriches the information content of the cyclic influence coefficient XH, so that it can more comprehensively reflect the impact of the classification results on the image features.

[0134] The beneficial effect of the cyclic influence coefficient XH calculated and output by this algorithm is that it can adaptively adjust the image features according to the classification results, thereby optimizing the image feature extraction process and improving the accuracy of classification. At the same time, through the cyclic influence mechanism, the cyclic influence coefficient XH can continuously iteratively update the image feature extraction index TQ, so that the classification system can continuously optimize its own performance according to the classification results, forming a virtuous circle. This cyclic influence mechanism not only optimizes the image feature extraction process, but also improves the accuracy and robustness of classification.

[0135] See also Figures 1 to 4 , the adjustment calculation formula based on the circulation influence coefficient XH is as follows:

[0136] TQ new =TQ+XH;

[0137] in:

[0138] TQ new is the new round of image feature extraction index, and the new round of image feature extraction index TQ new The reliability unit reflecting the correct classification of the image and the loop and optimization image feature extraction unit are introduced in sequence to form an iterative loop. The specific iterative loop adjustment and iterative stop methods are as follows:

[0139] If the cyclic influence coefficient XH is 0, it means that the classification result has no feedback effect on the image feature extraction index TQ. When combined with the new round of image feature extraction index TQ new Carry out a new round of iteration;

[0140] If the cyclic influence coefficient XH is a positive number, it reflects that the classification result has a positive feedback effect on the image feature extraction index TQ, and when the result of the cyclic influence coefficient XH continues to approach 1, the iteration is stopped. In addition, if the cyclic influence coefficient XH is a positive number and remains a positive value during the iteration process, the iteration is also stopped;

[0141] If the cyclic influence coefficient XH is negative, it means that the classification result has a negative feedback effect on the image feature extraction index TQ, so the classifier parameters should be checked and adjusted.

[0142] In this embodiment, through the cyclic influence mechanism, the cyclic influence coefficient XH obtained can continuously optimize the image feature extraction index TQ in the overall feature unit of the image, so that it can more accurately reflect the pathological information of the image. This optimization process not only improves the accuracy of feature extraction, but also enhances the classifier's recognition ability for different types of breast tissue pathology images. With the continuous optimization of the image feature extraction index TQ, the classifier can more accurately identify the key features in the breast tissue pathology image, thereby improving the accuracy of classification. This improvement in accuracy helps to reduce the risk of misdiagnosis and missed diagnosis, and provides strong support for the early diagnosis and treatment of breast cancer. In addition, the cyclic influence mechanism enables the classification system to adapt to different types of breast tissue pathology images, enhances the robustness and generalization ability of the system, and this improvement in robustness helps to ensure that the classification system can maintain stable performance when facing complex and changeable breast tissue pathology images.

[0143] Specifically, by calculating the loop influence coefficient XH, the feedback effect of the classification result on the image feature extraction index can be quantified. When the loop influence coefficient XH is negative, it indicates that the current classifier parameter setting is not reasonable, resulting in a deviation between the classification result and the actual image feature. At this time, according to the feedback of the loop influence coefficient XH, the classifier parameters are adjusted, including changing the decision threshold, increasing and decreasing the number of features, which can gradually optimize the performance of the classifier, thereby improving the classification accuracy. In the process of loop feedback, when the loop influence coefficient XH is positive and tends to 1, and maintains a positive value unchanged, it means that the feedback effect of the classification result on the image feature extraction index has stabilized. At this time, continuing to iterate cannot further significantly improve the classification accuracy, but will increase the consumption of computing resources. Therefore, setting these benchmarks for stopping iterations can avoid the waste of computing resources caused by excessive iterations and optimize the utilization efficiency of computing resources.

[0144] The loop feedback mechanism continuously adjusts the classifier parameters to adapt to different image features, which helps to enhance the robustness of the system. In practical applications, breast tissue pathology images have a variety of complex pathological features, and the loop feedback mechanism can gradually optimize the classifier's ability to recognize these features, allowing the system to more accurately identify different types of breast tissue pathology images, thereby improving the overall performance of the system;

[0145] In addition, the cyclic influence coefficient XH provides a clear basis for adjusting the classifier parameters. By calculating the cyclic influence coefficient XH, we can understand the feedback of the classification results under the current classifier parameter settings, so as to adjust the parameters in a targeted manner to optimize the classification performance. This data-driven parameter adjustment method is more scientific and efficient than the traditional trial and error method.

[0146] Setting a loop feedback benchmark not only helps to complete the current classification task, but also provides a direction for the continuous improvement of the algorithm. By analyzing the changing trend of the loop influence coefficient XH and the adjustment effect of the classifier parameters, we can gain an in-depth understanding of the performance of the algorithm in different situations, thereby discovering potential problems and improvement points. This helps to promote the continuous iteration and optimization of the algorithm and improve its performance in the breast tissue pathology image classification task.

[0147] In summary, the unit that comprehensively reflects the overall characteristics of the image, the unit that reflects the reliability of the correct classification of the image, and the cyclic and optimized image feature extraction unit each play an important role in the breast tissue pathology image classification method and system, and they are interconnected through the cyclic influence mechanism, which jointly improves the accuracy and efficiency of classification. This cyclic influence mechanism not only optimizes the image feature extraction process, but also improves the accuracy and robustness of classification, providing strong support for the early diagnosis and treatment of breast cancer. Setting the above-mentioned cyclic feedback benchmark for the cyclic influence coefficient XH has significant beneficial effects in the breast tissue pathology image classification method and system, including improving classification accuracy, optimizing computing resource utilization, enhancing system robustness, providing a basis for parameter adjustment, and promoting continuous improvement of the algorithm.

[0148] For example 2, please refer to Figures 1 to 4 , image gray value standard deviation BC H The calculation formula is as follows:

[0149]

[0150] H avg =(H1+H2+H3+......+H N ) / N;

[0151] H avg is the average gray value of pixels;

[0152] H1 is the grayscale value of the first pixel, H2 is the grayscale value of the second pixel, H3 is the grayscale value of the third pixel, N is the gray value of the Nth pixel;

[0153] Image texture eigenvalue standard deviation BC W The calculation formula is as follows:

[0154]

[0155] W avg =(W1+W2+W3+......+W N ) / N;

[0156] W avg is the average eigenvalue of pixel texture;

[0157] W1 is the first pixel texture eigenvalue, W2 is the second pixel texture eigenvalue, W3 is the third pixel texture eigenvalue, W N is the Nth pixel texture feature value;

[0158] The standard deviation of the number of abnormal cells in the image is BC BL The calculation formula is as follows:

[0159]

[0160] BL avg =(BL1+BL2+BL3+......+BL N ) / N;

[0161] BL1 is the number of abnormal cells in the first pixel, BL2 is the number of abnormal cells in the second pixel, BL3 is the number of abnormal cells in the third pixel, and BL N is the number of abnormal cells in the Nth pixel.

[0162] In this embodiment, first, the image gray value standard deviation BC H It is one of the statistics to measure the degree of data dispersion. It indicates the degree of deviation between the data point and the average value. In image processing, the standard deviation of the gray value can reflect the fluctuation of the image gray value, that is, the contrast and clarity of the image. The standard deviation of the image gray value BC H The larger the value, the greater the fluctuation of the image gray value, and the clearer the image. On the contrary, the standard deviation BC of the image gray value is H The smaller it is, the blurrier the image;

[0163] Image texture eigenvalue standard deviation BC W The standard deviation of the image gray value BC H Similarly, the standard deviation of image texture eigenvalues ​​BC WIt reflects the fluctuation of image texture features. The standard deviation of image texture feature value BC W The larger the value, the greater the fluctuation of the image texture characteristics, that is, the more complex the texture. On the contrary, the standard deviation BC of the image texture characteristic value is W The smaller it is, the simpler the texture;

[0164] The standard deviation of the number of abnormal cells in the image is BC BL It reflects the fluctuation of pathological features in the image. In the classification of breast pathological images, the standard deviation of the number of abnormal cells in the image is BC. BL The larger the value, the wider the distribution of pathological features and the greater the variation. Conversely, the standard deviation BC of the number of abnormal cells in the image BL The smaller it is, the more concentrated the distribution of pathological features is and the smaller the variation is, which helps the classification system to better identify and analyze pathological features.

[0165] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for classifying breast tissue pathology images, characterized in that: It includes a method for adaptively and iteratively adjusting image features according to classification results; The specific implementation steps are as follows: Step 1: using an image acquisition module to acquire images of breast tissue pathology; Step 2: Using the image processing module, the collected breast tissue pathology image is subjected to feature extraction processing, wherein the feature extraction includes gray value extraction, texture feature extraction, and abnormal cell number extraction of pathological features; Step 3: Based on feature extraction and using the image data calculation module, the output image feature extraction index TQ, classification confidence KX, and cyclic influence coefficient XH are calculated in sequence; Step 4: Based on the cyclic influence coefficient XH, iterative processing is performed using an iterative analysis module; The image data calculation module includes a unit for comprehensively reflecting the overall characteristics of the image, a unit for reflecting the reliability of the correct classification of the image, and a unit for cyclically and optimizing the image feature extraction.

2. A breast tissue pathology image classification method according to claim 1, characterized in that: The equipment used by the image acquisition module includes a microscope and a scanner; The equipment used in the image processing module includes medical image processing software and professional pathology analysis tools; The equipment used by the image data calculation module includes a high-performance computer; The equipment used by the iterative analysis module includes iterative control and analysis tools.

3. The method for classifying breast pathological images according to claim 2, characterized in that: The calculation formula that fully reflects the overall feature unit of the image is as follows: in: TQ is the image feature extraction index; N is the total number of pixels in the image; H i is the gray value of the i-th pixel; W i is the texture feature value of the i-th pixel; BL i is the number of abnormal cells in the i-th pixel.

4. The method for classifying breast pathological images according to claim 3, characterized in that: The i-th pixel texture feature value W i The calculation formula is as follows: h is the number of gray levels; a and b are grayscale values. a and b represent the grayscale values ​​of two pixels at a certain distance in the image, and the two pixels are adjacent; HG(a,b) is the value of the element (a,b) in the gray-level co-occurrence matrix, which represents the joint probability that two pixels at a certain distance in the image have gray-level values ​​a and b respectively; Among them, if the image is an 8-bit grayscale image, then the number of gray levels h is 256 (from 0 to 255). When calculating the grayscale co-occurrence matrix, it is necessary to introduce the grayscale values ​​of two pixels in the image that are a certain distance apart, namely a and b. Since the grayscale value is an integer from 0 to h-1, h-1 represents the maximum grayscale value.

5. The method for classifying breast tissue pathology images according to claim 3, characterized in that: The calculation formula reflecting the reliability unit of the image being correctly classified is as follows: KX=TQ×BC H -[(BL max -BL min ) / (TQ+SQRT(BC W ))]; in: KX is the classification confidence; BC H is the standard deviation of the image gray value; BL max is the maximum number of abnormal cells; BL min is the minimum number of abnormal cells; BC W is the standard deviation of the image texture eigenvalues.

6. A breast tissue pathology image classification method according to claim 5, characterized in that: The calculation formula of the cycle and optimization image feature extraction unit is as follows: XH=[(KX×(TQ+SQRT(BL avg ))) / (1+(BC BL / TQ) 2 )]-((KX-TQ) / N); in: XH is the circulation influence coefficient; BL avg is the average number of abnormal cells; BC BL is the standard deviation of the number of abnormal cells in the image.

7. A breast tissue pathology image classification method according to claim 6, characterized in that: The adjustment calculation formula based on the circulation influence coefficient XH is as follows: China new =TQ+XH; in: TQ new is the new round of image feature extraction index, and the new round of image feature extraction index TQ new The unit reflecting the reliability of correct classification of the image and the unit for extracting the loop and optimized image features are introduced in sequence to form an iterative loop. The specific method for adjusting the iterative loop and stopping the iteration is as follows: If the cyclic influence coefficient XH is 0, it means that the classification result has no feedback effect on the image feature extraction index TQ. When combined with the new round of image feature extraction index TQ new Carry out a new round of iteration; If the cyclic influence coefficient XH is a positive number, it reflects that the classification result has a positive feedback effect on the image feature extraction index TQ, and when the result of the cyclic influence coefficient XH continues to approach 1, the iteration is stopped. In addition, if the cyclic influence coefficient XH is a positive number and remains a positive value during the iteration process, the iteration is also stopped; If the cyclic influence coefficient XH is negative, it means that the classification result has a negative feedback effect on the image feature extraction index TQ, so the classifier parameters should be checked and adjusted.

8. The method for classifying breast pathological images according to claim 6, characterized in that: The image gray value standard deviation BC H The calculation formula is as follows: H avg =(H1+H2+H3+......+H N ) / N; H avg is the average gray value of pixels; H1 is the grayscale value of the first pixel, H2 is the grayscale value of the second pixel, H3 is the grayscale value of the third pixel, N is the gray value of the Nth pixel; The standard deviation BC of the image texture feature value W The calculation formula is as follows: W avg =(W1+W2+W3+......+W N ) / N; W avg is the average eigenvalue of pixel texture; W1 is the first pixel texture eigenvalue, W2 is the second pixel texture eigenvalue, W3 is the third pixel texture eigenvalue, W N is the Nth pixel texture feature value; The standard deviation BC of the number of abnormal cells in the image BL The calculation formula is as follows: <h2 style=";text-align:left;direction:ltr">BL<h2 style=";text-align:left;direction:ltr"> avg <h2 style=";text-align:left;direction:ltr"> (BL1+BL2+BL3+......+BL)<h2 style=";text-align:left;direction:ltr"> N <h2 style=";text-align:left;direction:ltr"> ) / N; BL1 is the number of abnormal cells in the first pixel, BL2 is the number of abnormal cells in the second pixel, BL3 is the number of abnormal cells in the third pixel, and BL N is the number of abnormal cells in the Nth pixel.

9. The image classification system used in the breast tissue pathology image classification method according to claim 1, characterized in that: include: Image acquisition module: responsible for collecting breast tissue pathology images; Image processing module: responsible for feature extraction of abnormal cell numbers including grayscale value, texture features and pathological features: Image data calculation module: responsible for calculating the output image feature extraction index TQ, classification confidence KX, and cyclic influence coefficient XH in sequence; Iterative analysis module: responsible for iteration based on the cycle influence coefficient XH.

Images