System for realizing automatic grading of breast cancer cell nucleuses by using deep learning technology

Abstract

The invention provides a system for realizing automatic grading of breast cancer cell nucleuses by using a deep learning technology, a historical breast cancer slide is scanned through a scanner to obtain a historical full-slice image, and an initial artificial intelligence model is trained based on annotation data of a high-grade pathologist on the historical full-slice image to obtain a breast cancer cell nucleus classification model; the method comprises the following steps: obtaining an automatic grading model, scanning a to-be-detected breast cancer slide through a scanner to obtain a to-be-detected full-slice image, and inputting the to-be-detected full-slice image into the automatic grading model to obtain a breast cancer cell nucleus grading result of the to-be-detected full-slice image. The automatic grading model is updated based on the difference between the breast cancer cell nucleus grading result of the full slice image to be detected and the actual application result, automatic interpretation is achieved, the working efficiency of breast cancer cell nucleus grade interpretation is improved, over-diagnosis and insufficient diagnosis are prevented, large-scale image processing work is completed more quickly, and the detection efficiency of breast cancer cell nucleus grade interpretation is improved. And the analysis efficiency and the diagnosis accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a system for automatically grading breast cancer cell nuclei using deep learning technology. Background Art

[0002] Breast cancer is one of the most common malignant tumors in women worldwide, posing a threat to their physical and mental health. Early diagnosis and treatment are key factors in determining the prognosis of breast cancer. To objectively evaluate the malignancy of breast cancer, a semi-quantitative grading system has been proposed, in which nuclear pleomorphism is classified into three levels: mild, moderate, and severe. Traditional grading of nuclear pleomorphism in breast cancer relies primarily on the subjective judgment of pathologists. This method lacks quantitative analysis of nuclear pleomorphism and is not only time-consuming and labor-intensive, but also suffers from low intra- and inter-observer reproducibility, making it difficult to meet growing clinical needs.

[0003] The existing breast cancer cell nuclear grading system is mainly based on the subjective interpretation of breast cancer stained sections by pathologists under a microscope. Its main shortcomings are as follows:

[0004] Manual interpretation by pathologists under a microscope lacks quantitative interpretation standards, resulting in poor reproducibility and low consistency in interpretations by different pathologists.

[0005] Manual interpretation by pathologists is easily affected by subjective conditions and has a certain degree of subjectivity;

[0006] Manual interpretation by pathologists is not efficient and they can easily become fatigued after reading for a long time, thus affecting the accuracy of the interpretation. Summary of the Invention

[0007] The present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, so as to solve the problems raised in the background technology.

[0008] A system for automatically grading breast cancer cell nuclei using deep learning technology, comprising:

[0009] A scanning module is used to scan historical breast cancer slides through a scanner to obtain historical full-slice images;

[0010] The model training module is used to train the initial AI model based on the annotation data of historical full-slide images by senior pathologists to obtain an automatic grading model;

[0011] A grading module is used to scan a breast cancer slide to be detected by a scanner to obtain a full-slice image to be detected, input the full-slice image to be detected into an automatic grading model, and obtain a breast cancer cell nucleus grading result for the full-slice image to be detected;

[0012] The model updating module is used to update the automatic grading model based on the difference between the breast cancer cell nucleus grading result of the full-slice image to be detected and the actual application result.

[0013] Preferably, the scanning module includes:

[0014] a determining unit, configured to determine a relative position and a relative direction of the historical breast cancer slide relative to the scanner based on a scanning requirement;

[0015] The scanning unit is used to control the historical breast cancer slide to be scanned by a scanner according to the relative position and relative direction to obtain a historical full-slice image.

[0016] Preferably, the model training module includes:

[0017] a labeling unit, configured to respectively obtain first labeling data and second labeling data of historical full-slide images by two senior pathologists;

[0018] a processing unit, configured to process the first annotated data based on an annotation difference between the first annotated data and the second annotated data to obtain annotated data for the historical full-slice image;

[0019] The model building unit is used to use the labeled data as training samples to train the artificial intelligence model to obtain an automatic grading model.

[0020] Preferably, the processing unit includes:

[0021] a difference determining unit, configured to obtain a marked difference between the first marked data and the second marked data, and obtain a first position and a first level, and a second position and a second level in a standard difference;

[0022] a preference determining unit, configured to obtain a position difference between the first position and the second position, and a level difference between the first level and the second level, and determine a subjective labeling preference for the first labeling data based on the position difference and the level difference;

[0023] A correction unit is used to correct the subjective annotation preference based on a preset correction table, and perform correction processing on the first annotation data according to the correction result to obtain annotation data for the historical full-slice image.

[0024] Preferably, the model building unit includes:

[0025] a division unit, configured to divide the annotated data based on the number of annotations, the position of annotations, and the level of annotations, to obtain a number of annotation data groups, a position of annotation data groups, and a level of annotation data groups;

[0026] A first training unit is used to train the artificial intelligence model based on the quantity labeled data set, obtain multiple sets of quantity training results, and determine the artificial intelligence model's ability to recognize quantity based on the quantity training results;

[0027] The second training unit is used to train the artificial intelligence model based on the location annotation data group to obtain multiple sets of location training results, and determine the artificial intelligence model's ability to recognize locations based on the location training results;

[0028] The third training unit is used to train the artificial intelligence model based on the level-labeled data group, obtain multiple sets of level training results, and determine the level recognition ability of the artificial intelligence model based on the level training results;

[0029] a parameter determination unit, configured to determine a first adjustment parameter, a second adjustment parameter, and a third adjustment parameter for the artificial intelligence model based on the artificial intelligence model's ability to recognize quantity, location, and level, and to determine a comprehensive adjustment parameter based on a correlation among the first adjustment parameter, the second adjustment parameter, and the third adjustment parameter;

[0030] an adjustment unit, configured to adjust the trained artificial intelligence model based on the comprehensive adjustment parameters to obtain an initial classification model;

[0031] The verification unit is used to randomly extract labeled data to verify the initial classification model, and obtain the automatic classification model after the verification is passed.

[0032] Preferably, the parameter determination unit includes:

[0033] an operation determination unit, configured to determine operation merging data between the first adjustment parameter, the second adjustment parameter, and the third adjustment parameter based on the correlation between the first adjustment parameter, the second adjustment parameter, and the third adjustment parameter;

[0034] The operation processing unit is used to perform operation processing on the first adjustment parameter, the second adjustment parameter and the third adjustment parameter based on the operation combined data to determine the comprehensive adjustment parameter.

[0035] Preferably, the grading module includes:

[0036] A slide scanning unit is used to scan the breast cancer slide to be detected through a scanner according to a preset position and direction to obtain an image of the entire slide to be detected;

[0037] The cell nucleus grading unit is used to input the full slice image to be detected into the automatic grading model, determine the grading labeling information of the breast cancer cell nuclei in the full slice image, and obtain the breast cancer cell nucleus grading result of the full slice image to be detected based on the grading labeling information.

[0038] Preferably, the model updating module includes:

[0039] a result acquisition unit, configured to acquire a breast cancer cell nuclear grading result of the full-slice image to be detected, and to acquire a situation in which the breast cancer cell nuclear grading result is actually applied, and to determine an actual application result;

[0040] a marking unit, configured to obtain a grading difference feature of the breast cancer cell nucleus grading result from the actual application result, and mark the whole slice image to be detected based on the grading difference feature to obtain a marked image;

[0041] a training unit, configured to train the automatic classification model using a preset number of labeled images as second training samples to obtain an intermediate classification model;

[0042] a difference determining unit, configured to obtain model difference data between the automatic grading model and the intermediate grading model, and to obtain a model bias difference, a model variance difference, and a model noise difference from the model difference data;

[0043] a difference analysis unit, configured to determine an increase in model complexity based on the model bias difference, determine adjustment parameters for a model regularization technique based on the model variance difference, and determine an optimization value for model data processing based on the model noise difference;

[0044] An updating unit is configured to update the automatic grading model based on the added value of the model complexity, the adjustment parameters of the model regularization technique, and the optimized value of the model data processing.

[0045] Preferably, the marking unit comprises:

[0046] a feature analysis unit, configured to obtain a grading difference feature of the breast cancer cell nucleus grading result from the actual application result, and obtain a difference type and a difference degree of the grading difference feature;

[0047] The image marking unit is used to perform a first marking on the full-slice image to be detected based on the difference type and a second marking on the full-slice image to be detected based on the difference degree to obtain a marked image.

[0048] Preferably, the updating unit includes:

[0049] an update determination unit, configured to determine a first update parameter for the model based on the increase in model complexity, determine a second update parameter for the model based on an adjustment parameter of a model regularization technique, and determine a third update parameter for the model based on an optimized value of model data processing;

[0050] The model updating unit is configured to determine a target update parameter of the automatic grading model based on a parameter relationship among the first update parameter, the second update parameter and the third update parameter, and to update the automatic grading model based on the target update parameter.

[0051] Compared with the prior art, the present invention achieves the following beneficial effects: historical breast cancer slides are scanned with a scanner to obtain historical full-slide images; an initial artificial intelligence model is trained based on the annotation data of the historical full-slide images by senior pathologists to obtain an automatic grading model; a breast cancer slide to be tested is scanned with a scanner to obtain a full-slide image to be tested; the full-slide image to be tested is input into the automatic grading model to obtain a breast cancer cell nucleus grading result for the full-slide image to be tested; the automatic grading model is updated based on the difference between the breast cancer cell nucleus grading result of the full-slide image to be tested and the actual application result, thereby achieving automated interpretation, improving the efficiency of breast cancer cell nucleus grading, saving manpower and material resources; the cell nucleus grading results obtained by model interpretation are relatively quantitative, improving objectivity and repeatability, and preventing overdiagnosis and underdiagnosis; the model recognizes breast cancer full-slide images to achieve automated identification and grading of cell nuclei; the model automatically identifies the location and grade of cell nuclei in breast full-slide images, and directly obtains the breast cancer cell nucleus result; ultimately, the model can assist pathologists in automatically extracting breast cancer image features, completing large-scale image processing tasks more quickly, and improving analysis efficiency and diagnostic accuracy.

[0052] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in this application document.

[0053] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0055] Figure 1 This is a structural diagram of a system for automatically grading breast cancer cell nuclei using deep learning technology in an embodiment of the present invention;

[0056] Figure 2 4 is a structural diagram of the model training module described in an embodiment of the present invention;

[0057] Figure 3 4 is a structural diagram of the grading module in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0059] Example 1:

[0060] The embodiment of the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology. Figure 1 Shown, including:

[0061] A scanning module is used to scan historical breast cancer slides through a scanner to obtain historical full-slice images;

[0062] The model training module is used to train the initial AI model based on the annotation data of historical full-slide images by senior pathologists to obtain an automatic grading model;

[0063] A grading module is used to scan a breast cancer slide to be detected by a scanner to obtain a full-slice image to be detected, input the full-slice image to be detected into an automatic grading model, and obtain a breast cancer cell nucleus grading result for the full-slice image to be detected;

[0064] The model updating module is used to update the automatic grading model based on the difference between the breast cancer cell nucleus grading result of the full-slice image to be detected and the actual application result.

[0065] In this embodiment, the whole-slice image is a high-resolution digital image acquired by scanning with a fully automatic microscope or an optical magnification system, and is stitched and processed by a computer with high precision and seamless multi-view to obtain a multi-level visual image.

[0066] In this embodiment, the annotation data of the historical full-slice images by the senior pathologist specifically includes annotation of the historical full-slice images at the cell nucleus level.

[0067] In this embodiment, the actual application result of the full-slice image to be detected is the inaccuracy caused by applying the breast cancer cell nucleus grading result to the actual process.

[0068] The beneficial effects of the above design scheme are as follows: historical breast cancer slides are scanned with a scanner to obtain historical full-slice images, an initial artificial intelligence model is trained based on the annotation data of the historical full-slice images by senior pathologists to obtain an automatic grading model, a breast cancer slide to be tested is scanned with a scanner to obtain a full-slice image to be tested, the full-slice image to be tested is input into the automatic grading model, and a breast cancer cell nucleus grading result for the full-slice image to be tested is obtained; based on the difference between the breast cancer cell nucleus grading result of the full-slice image to be tested and the actual application result, the automatic grading model is updated to achieve automated interpretation, improve the efficiency of breast cancer cell nucleus grading, save manpower and material resources, the cell nucleus grading results obtained by model interpretation are relatively quantitative, improve objectivity and repeatability, and prevent overdiagnosis and underdiagnosis, the model recognizes breast cancer full-slice images to achieve automated identification and grading of cell nuclei, the model automatically identifies the location and grade of cell nuclei in breast full-slice images, and directly obtains the breast cancer cell nucleus result, which can ultimately assist pathologists in automatically extracting breast cancer image features, completing large-scale image processing tasks more quickly, and improving analysis efficiency and diagnostic accuracy.

[0069] Example 2:

[0070] Based on Example 1, this embodiment of the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, wherein the scanning module includes:

[0071] a determining unit, configured to determine a relative position and a relative direction of the historical breast cancer slide relative to the scanner based on a scanning requirement;

[0072] The scanning unit is used to control the historical breast cancer slide to be scanned by a scanner according to the relative position and relative direction to obtain a historical full-slice image.

[0073] The beneficial effect of the above design scheme is: by determining the relative position and relative direction of the historical breast cancer slide relative to the scanner based on the scanning requirements, the historical breast cancer slide is controlled to be scanned through the scanner according to the relative position and relative direction, and the historical full-slice image is obtained, providing a training sample basis for the training of the artificial intelligence model.

[0074] Example 3:

[0075] Based on Example 1, the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, such as Figure 2 As shown, the model training module includes:

[0076] a labeling unit, configured to respectively obtain first labeling data and second labeling data of historical full-slide images by two senior pathologists;

[0077] a processing unit, configured to process the first annotated data based on an annotation difference between the first annotated data and the second annotated data to obtain annotated data for the historical full-slice image;

[0078] The model building unit is used to use the labeled data as training samples to train the artificial intelligence model to obtain an automatic grading model.

[0079] The beneficial effects of the above design scheme are as follows: by respectively obtaining first and second annotation data of historical full-slice images from two senior pathologists, the first annotation data is processed based on the annotation difference between the first and second annotation data to obtain annotation data for the historical full-slice images, thereby ensuring the accuracy of the manually annotated data and providing high-quality training samples for model training; by using the annotated data as training samples, the artificial intelligence model is trained to obtain an automatic grading model; the cell nucleus grading results obtained by the model interpretation are relatively quantitative, thereby improving objectivity and repeatability, and preventing overdiagnosis and underdiagnosis; the model recognizes breast cancer full-slice images to realize automatic identification and grading of cell nuclei; the model automatically recognizes the position and grade of cell nuclei in breast full-slice images, and directly obtains the results of breast cancer cell nuclei; ultimately, it can assist pathologists in automatically extracting breast cancer image features, completing large-scale image processing work more quickly, and improving analysis efficiency and diagnostic accuracy.

[0080] Example 4:

[0081] Based on Example 3, this embodiment of the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, wherein the processing unit includes:

[0082] a difference determining unit, configured to obtain a marked difference between the first marked data and the second marked data, and obtain a first position and a first level, and a second position and a second level in a standard difference;

[0083] a preference determining unit, configured to obtain a position difference between the first position and the second position, and a level difference between the first level and the second level, and determine a subjective labeling preference for the first labeling data based on the position difference and the level difference;

[0084] A correction unit is used to correct the subjective annotation preference based on a preset correction table, and perform correction processing on the first annotation data according to the correction result to obtain annotation data for the historical full-slice image.

[0085] In this embodiment, the preset correction table is designed in advance based on historical experience, and can correct the influence of human subjectivity.

[0086] The beneficial effects of the above design scheme are: by obtaining the annotation difference between the first annotation data and the second annotation data, obtaining the first position and the first level, as well as the second position and the second level in the standard difference, obtaining the position difference between the first position and the second position, and the level difference between the first level and the second level, and based on the position difference and the level difference, determining the subjective annotation preference for the first annotation data, correcting the subjective annotation preference based on a preset correction table, and correcting the first annotation data according to the correction result, obtaining the annotation data of the historical full-slice image, ensuring the accuracy of the manually annotated data, and providing high-quality training samples for model training.

[0087] Example 5:

[0088] Based on Example 3, this embodiment of the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, wherein the model building unit includes:

[0089] a division unit, configured to divide the annotated data based on the number of annotations, the position of annotations, and the level of annotations, to obtain a number of annotation data groups, a position of annotation data groups, and a level of annotation data groups;

[0090] A first training unit is used to train the artificial intelligence model based on the quantity labeled data set, obtain multiple sets of quantity training results, and determine the artificial intelligence model's ability to recognize quantity based on the quantity training results;

[0091] The second training unit is used to train the artificial intelligence model based on the location annotation data group to obtain multiple sets of location training results, and determine the artificial intelligence model's ability to recognize locations based on the location training results;

[0092] The third training unit is used to train the artificial intelligence model based on the level-labeled data group, obtain multiple sets of level training results, and determine the level recognition ability of the artificial intelligence model based on the level training results;

[0093] a parameter determination unit, configured to determine a first adjustment parameter, a second adjustment parameter, and a third adjustment parameter for the artificial intelligence model based on the artificial intelligence model's ability to recognize quantity, location, and level, and to determine a comprehensive adjustment parameter based on a correlation among the first adjustment parameter, the second adjustment parameter, and the third adjustment parameter;

[0094] an adjustment unit, configured to adjust the trained artificial intelligence model based on the comprehensive adjustment parameters to obtain an initial classification model;

[0095] The verification unit is used to randomly extract labeled data to verify the initial classification model, and obtain the automatic classification model after the verification is passed.

[0096] In this embodiment, the quantity annotation data set is that the annotation quantity of each group of annotation data is within a certain range, the position annotation data set is that the annotation position of each group of annotation data is within a certain area, and the level annotation data set is that the annotation level of each group of annotation data is within a certain range.

[0097] In this embodiment, the quantity corresponds to the first adjustment parameter, the position corresponds to the second adjustment parameter, and the level corresponds to the third adjustment parameter.

[0098] In this embodiment, the initial classification model is verified by randomly sampling labeled data, and if the verification fails, the model parameters need to be adjusted again.

[0099] The beneficial effects of the above design scheme are: by using the number of annotations on the whole slice image, the annotation position and the annotation level as the three factors of model training for model training and verification, an automatic grading model is obtained by training separately and finally randomly sampling and verifying. The cell nucleus grading results obtained by the model interpretation are relatively quantitative, which improves objectivity and repeatability, prevents overdiagnosis and underdiagnosis, and realizes automatic identification and grading of cell nuclei through model recognition of breast cancer whole slice images. The model automatically identifies the position and level of cell nuclei in breast whole slice images, and directly obtains the results of breast cancer cell nuclei. Ultimately, it can assist pathologists in automatically extracting breast cancer image features, complete large-scale image processing work more quickly, and improve analysis efficiency and diagnostic accuracy.

[0100] Example 6:

[0101] Based on Example 5, this embodiment of the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, wherein the parameter determination unit includes:

[0102] an operation determination unit, configured to determine operation merging data between the first adjustment parameter, the second adjustment parameter, and the third adjustment parameter based on the correlation between the first adjustment parameter, the second adjustment parameter, and the third adjustment parameter;

[0103] The operation processing unit is used to perform operation processing on the first adjustment parameter, the second adjustment parameter and the third adjustment parameter based on the operation combined data to determine the comprehensive adjustment parameter.

[0104] In this embodiment, the operation to merge data is, for example, addition, subtraction, multiplication, and division of parameters.

[0105] The beneficial effect of the above design scheme is: by determining the operational merge data between the first adjustment parameter, the second adjustment parameter and the third adjustment parameter based on the correlation between the first adjustment parameter, the second adjustment parameter and the third adjustment parameter, and based on the operational merge data, performing operational processing on the first adjustment parameter, the second adjustment parameter and the third adjustment parameter to determine the comprehensive adjustment parameter, thereby facilitating the adjustment of the model parameters.

[0106] Example 7:

[0107] Based on Example 1, the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, such as Figure 3 As shown, the grading module includes:

[0108] A slide scanning unit is used to scan the breast cancer slide to be detected through a scanner according to a preset position and direction to obtain an image of the entire slide to be detected;

[0109] The cell nucleus grading unit is used to input the full slice image to be detected into the automatic grading model, determine the grading labeling information of the breast cancer cell nuclei in the full slice image, and obtain the breast cancer cell nucleus grading result of the full slice image to be detected based on the grading labeling information.

[0110] The beneficial effects of the above design scheme are as follows: by scanning the breast cancer slide to be tested with a scanner according to a preset position and direction, a full-slice image to be tested is obtained, the full-slice image to be tested is input into an automatic grading model, grading labeling information for the breast cancer cell nuclei in the full-slice image is determined, and based on the grading labeling information, a breast cancer cell nucleus grading result for the full-slice image to be tested is obtained, thereby achieving automated interpretation, improving the efficiency of breast cancer cell nucleus grading, saving manpower and material resources, and the cell nucleus grading results obtained by model interpretation are relatively quantitative, thereby improving objectivity and repeatability.

[0111] Example 8:

[0112] Based on Example 1, this embodiment of the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, wherein the model updating module includes:

[0113] a result acquisition unit, configured to acquire a breast cancer cell nuclear grading result of the full-slice image to be detected, and to acquire a situation in which the breast cancer cell nuclear grading result is actually applied, and to determine an actual application result;

[0114] a marking unit, configured to obtain a grading difference feature of the breast cancer cell nucleus grading result from the actual application result, and mark the whole slice image to be detected based on the grading difference feature to obtain a marked image;

[0115] a training unit, configured to train the automatic classification model using a preset number of labeled images as second training samples to obtain an intermediate classification model;

[0116] a difference determining unit, configured to obtain model difference data between the automatic grading model and the intermediate grading model, and to obtain a model bias difference, a model variance difference, and a model noise difference from the model difference data;

[0117] a difference analysis unit, configured to determine an increase in model complexity based on the model bias difference, determine adjustment parameters for a model regularization technique based on the model variance difference, and determine an optimization value for model data processing based on the model noise difference;

[0118] An updating unit is configured to update the automatic grading model based on the added value of the model complexity, the adjustment parameters of the model regularization technique, and the optimized value of the model data processing.

[0119] The beneficial effects of the above design scheme are: by updating the automatic grading model based on the difference between the breast cancer cell nucleus grading results of the full-slice image to be tested and the actual application results, considering the three aspects of model query, variance and noise to determine the update parameters, the automatic grading model can be updated in real time as the breast cancer cell nucleus changes, ensuring that the updated automatic grading model can improve the accuracy of the automatic grading of cell nuclei, and ultimately assisting pathologists in automatically extracting breast cancer image features, completing large-scale image processing tasks more quickly, and improving analysis efficiency and diagnostic accuracy.

[0120] Example 9:

[0121] Based on Example 8, this embodiment of the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, wherein the labeling unit includes:

[0122] a feature analysis unit, configured to obtain a grading difference feature of the breast cancer cell nucleus grading result from the actual application result, and obtain a difference type and a difference degree of the grading difference feature;

[0123] The image marking unit is configured to perform a first marking on the full-slice image to be detected based on the difference type and a second marking on the full-slice image to be detected based on the difference degree to obtain a marked image.

[0124] The beneficial effect of the above design scheme is: by obtaining the grading difference characteristics of the breast cancer cell nucleus grading results from the actual application results, obtaining the difference type and difference degree of the grading difference characteristics, performing a first labeling on the full slice image to be detected based on the difference type, and performing a second labeling on the full slice image to be detected based on the difference degree, obtaining a labeled image, and providing accurate training samples for model training.

[0125] Example 10:

[0126] Based on Example 8, this embodiment of the present invention provides a system for automatically grading breast cancer cell nuclei using deep learning technology, wherein the updating unit includes:

[0127] an update determination unit, configured to determine a first update parameter for the model based on the increase in model complexity, determine a second update parameter for the model based on an adjustment parameter of a model regularization technique, and determine a third update parameter for the model based on an optimized value of model data processing;

[0128] The model updating unit is configured to determine a target update parameter of the automatic grading model based on a parameter relationship among the first update parameter, the second update parameter and the third update parameter, and to update the automatic grading model based on the target update parameter.

[0129] The beneficial effects of the above design scheme are: by determining the first update parameter of the model based on the increased value of the model complexity, determining the second update parameter of the model based on the adjustment parameter of the model regularization technology, determining the third update parameter of the model based on the optimized value of the model data processing, determining the target update parameter of the automatic grading model based on the parameter relationship between the first update parameter, the second update parameter and the third update parameter, and updating the automatic grading model based on the target update parameter, the automatic grading model is updated in real time as the breast cancer cell nucleus changes, ensuring that the updated automatic grading model can improve the accuracy of the automatic grading of the cell nucleus.

[0130] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of this application document and its equivalents, the present invention is intended to include these modifications and variations.

Claims

1. A system for automatically grading breast cancer cell nuclei using deep learning technology, characterized in that: include: A scanning module is used to scan historical breast cancer slides through a scanner to obtain historical full-slice images; The model training module is used to train the initial AI model based on the annotation data of historical full-slide images by senior pathologists to obtain an automatic grading model; A grading module is used to scan a breast cancer slide to be detected by a scanner to obtain a full-slice image to be detected, input the full-slice image to be detected into an automatic grading model, and obtain a breast cancer cell nucleus grading result for the full-slice image to be detected; The model updating module is used to update the automatic grading model based on the difference between the breast cancer cell nucleus grading result of the full-slice image to be detected and the actual application result.

2. A system for automatically grading breast cancer cell nuclei using deep learning technology according to claim 1, characterized in that: The scanning module includes: a determining unit, configured to determine a relative position and a relative direction of the historical breast cancer slide relative to the scanner based on a scanning requirement; The scanning unit is used to control the historical breast cancer slide to be scanned by a scanner according to the relative position and relative direction to obtain a historical full-slice image.

3. The system for realizing automatic grading of breast cancer cell nuclei using deep learning technology according to claim 1, characterized in that: The model training module includes: a labeling unit, configured to respectively obtain first labeling data and second labeling data of historical full-slide images by two senior pathologists; a processing unit, configured to process the first annotated data based on an annotation difference between the first annotated data and the second annotated data to obtain annotated data for the historical full-slice image; The model building unit is used to use the labeled data as training samples to train the artificial intelligence model to obtain an automatic grading model.

4. The system for realizing automatic grading of breast cancer cell nuclei using deep learning technology according to claim 3, characterized in that: The processing unit includes: a difference determining unit, configured to obtain a marked difference between the first marked data and the second marked data, and obtain a first position and a first level, and a second position and a second level in a standard difference; a preference determining unit, configured to obtain a position difference between the first position and the second position, and a level difference between the first level and the second level, and determine a subjective labeling preference for the first labeling data based on the position difference and the level difference; A correction unit is used to correct the subjective annotation preference based on a preset correction table, and perform correction processing on the first annotation data according to the correction result to obtain annotation data for the historical full-slice image.

5. The system for realizing automatic grading of breast cancer cell nuclei using deep learning technology according to claim 3, characterized in that: The model building unit includes: a division unit, configured to divide the annotated data based on the number of annotations, the position of annotations, and the level of annotations, to obtain a number of annotation data groups, a position of annotation data groups, and a level of annotation data groups; A first training unit is used to train the artificial intelligence model based on the quantity labeled data set, obtain multiple sets of quantity training results, and determine the artificial intelligence model's ability to recognize quantity based on the quantity training results; The second training unit is used to train the artificial intelligence model based on the location annotation data group to obtain multiple sets of location training results, and determine the artificial intelligence model's ability to recognize locations based on the location training results; The third training unit is used to train the artificial intelligence model based on the level-labeled data group, obtain multiple sets of level training results, and determine the level recognition ability of the artificial intelligence model based on the level training results; a parameter determination unit, configured to determine a first adjustment parameter, a second adjustment parameter, and a third adjustment parameter for the artificial intelligence model based on the artificial intelligence model's ability to recognize quantity, location, and level, and to determine a comprehensive adjustment parameter based on a correlation among the first adjustment parameter, the second adjustment parameter, and the third adjustment parameter; an adjustment unit, configured to adjust the trained artificial intelligence model based on the comprehensive adjustment parameters to obtain an initial classification model; The verification unit is used to randomly extract labeled data to verify the initial classification model, and obtain the automatic classification model after the verification is passed.

6. The system for realizing automatic grading of breast cancer cell nuclei using deep learning technology according to claim 5, characterized in that: The parameter determination unit includes: an operation determination unit, configured to determine operation merging data between the first adjustment parameter, the second adjustment parameter, and the third adjustment parameter based on the correlation between the first adjustment parameter, the second adjustment parameter, and the third adjustment parameter; The operation processing unit is used to perform operation processing on the first adjustment parameter, the second adjustment parameter and the third adjustment parameter based on the operation combined data to determine the comprehensive adjustment parameter.

7. The system for realizing automatic grading of breast cancer cell nuclei using deep learning technology according to claim 1, characterized in that: The grading module includes: A slide scanning unit is used to scan the breast cancer slide to be detected through a scanner according to a preset position and direction to obtain an image of the entire slide to be detected; The cell nucleus grading unit is used to input the full slice image to be detected into the automatic grading model, determine the grading labeling information of the breast cancer cell nuclei in the full slice image, and obtain the breast cancer cell nucleus grading result of the full slice image to be detected based on the grading labeling information.

8. The system for realizing automatic grading of breast cancer cell nuclei using deep learning technology according to claim 1, characterized in that: The model updating module includes: a result acquisition unit, configured to acquire a breast cancer cell nuclear grading result of the full-slice image to be detected, and to acquire a situation in which the breast cancer cell nuclear grading result is actually applied, and to determine an actual application result; a marking unit, configured to obtain a grading difference feature of the breast cancer cell nucleus grading result from the actual application result, and mark the whole slice image to be detected based on the grading difference feature to obtain a marked image; a training unit, configured to train the automatic classification model using a preset number of labeled images as second training samples to obtain an intermediate classification model; a difference determining unit, configured to obtain model difference data between the automatic grading model and the intermediate grading model, and to obtain a model bias difference, a model variance difference, and a model noise difference from the model difference data; a difference analysis unit, configured to determine an increase in model complexity based on the model bias difference, determine adjustment parameters for a model regularization technique based on the model variance difference, and determine an optimization value for model data processing based on the model noise difference; An updating unit is configured to update the automatic grading model based on the added value of the model complexity, the adjustment parameters of the model regularization technique, and the optimized value of the model data processing.

9. The system for realizing automatic grading of breast cancer cell nuclei using deep learning technology according to claim 8, characterized in that: The marking unit comprises: a feature analysis unit, configured to obtain a grading difference feature of the breast cancer cell nucleus grading result from the actual application result, and obtain a difference type and a difference degree of the grading difference feature; The image marking unit is configured to perform a first marking on the full-slice image to be detected based on the difference type and a second marking on the full-slice image to be detected based on the difference degree to obtain a marked image.

10. The system for realizing automatic grading of breast cancer cell nuclei using deep learning technology according to claim 8, characterized in that: The updating unit includes: an update determination unit, configured to determine a first update parameter for the model based on the increase in model complexity, determine a second update parameter for the model based on an adjustment parameter of a model regularization technique, and determine a third update parameter for the model based on an optimized value of model data processing; The model updating unit is configured to determine a target update parameter of the automatic grading model based on a parameter relationship among the first update parameter, the second update parameter and the third update parameter, and to update the automatic grading model based on the target update parameter.

Images